{ "meta": { "name": "AI Models Directory", "url": "https://guptadeepak.com/ai-models/", "description": "AI Models is a facts-only reference for the major language models from OpenAI, Anthropic, Google, Meta, Mistral, DeepSeek, xAI, Alibaba, Amazon, and Cohere. Every model card records the exact API model ID strings, release date, lifecycle status and announced sunset date, context window, license, whether the provider trains on API data by default, retention window and zero-data-retention availability, residency options, HIPAA BAA, SOC 2 and FedRAMP status, which surfaces serve the model, and a dated pricing snapshot with a link to the provider's own pricing page. It publishes no rankings.", "license": "https://creativecommons.org/licenses/by/4.0/", "attribution": "Deepak Gupta, guptadeepak.com", "generated": "2026-09-08", "modelCount": 66 }, "providers": [ { "id": "anthropic", "name": "Anthropic", "ownedDomains": [ "anthropic.com", "claude.com" ], "homepage": "https://www.anthropic.com/", "summary": "Claude model family. Publishes a model lifecycle with active, legacy, deprecated, and retired states, and commits to at least 60 days of notice before retiring a publicly released model." }, { "id": "openai", "name": "OpenAI", "ownedDomains": [ "openai.com" ], "homepage": "https://openai.com/", "summary": "GPT and o-series model families. Publishes a deprecations page and commits to at least 6 months of notice for generally available models." }, { "id": "google", "name": "Google", "ownedDomains": [ "google.com", "google.dev", "googleblog.com" ], "homepage": "https://ai.google.dev/", "summary": "Gemini model family, served through the first-party Gemini Developer API. Publishes a deprecation and shutdown schedule with an announced replacement model for each retiring ID." }, { "id": "meta", "name": "Meta", "ownedDomains": [ "meta.com", "llama.com" ], "homepage": "https://ai.meta.com/", "summary": "Llama open-weight model family. Meta releases weights and model cards but does not operate a metered first-party hosted API with a published per-token price; hosting and pricing are left to third-party inference providers." }, { "id": "mistral", "name": "Mistral AI", "ownedDomains": [ "mistral.ai" ], "homepage": "https://mistral.ai/", "summary": "Mistral, Ministral, and Codestral model families, served through La Plateforme. Publishes a models overview page carrying both current specs and a legacy deprecation and retirement table." }, { "id": "deepseek", "name": "DeepSeek", "ownedDomains": [ "deepseek.com" ], "homepage": "https://www.deepseek.com/", "summary": "DeepSeek V-series open-weight model family. Runs a standing peak and off-peak pricing schedule on its API rather than a single flat rate." }, { "id": "xai", "name": "xAI", "ownedDomains": [ "x.ai" ], "homepage": "https://x.ai/", "summary": "Grok model family. Publishes a release-notes changelog used as the lifecycle and retirement source, with a public migration guide for each retired model slug's replacement." }, { "id": "qwen", "name": "Qwen (Alibaba Cloud)", "ownedDomains": [ "alibabacloud.com", "aliyun.com" ], "homepage": "https://www.alibabacloud.com/help/en/model-studio/", "summary": "Qwen model family, served through Alibaba Cloud Model Studio. Publishes a dated model-release changelog and a snapshot-vs-mainline deprecation notice policy (30 days for dated snapshots, 3 months for mainline models)." }, { "id": "amazon", "name": "Amazon", "ownedDomains": [ "aws.amazon.com" ], "homepage": "https://aws.amazon.com/bedrock/", "summary": "Nova model family, served exclusively through Amazon Bedrock. Every model card publishes an explicit lifecycle state (Active, Legacy, or End-of-Life) with an EOL-no-sooner-than date and a Legacy notice period." }, { "id": "cohere", "name": "Cohere", "ownedDomains": [ "cohere.com" ], "homepage": "https://cohere.com/", "summary": "Command model family. Publishes a dated deprecations page naming each shutdown date and recommended replacement, and a public FAQ with per-model per-token pricing for its generally available models." } ], "models": [ { "slug": "claude-fable-5-1", "providerId": "anthropic", "displayName": "Claude Fable 5.1", "apiModelIds": [ "claude-fable-5-1" ], "releasedOn": "2026-09-01", "family": "claude-fable", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "contextWindow": 1000000, "maxOutputTokens": 128000, "license": "proprietary", "lifecycle": { "status": "active", "sunsetNotBefore": "2027-09-01", "verifiedAt": "2026-09-08", "source": "https://platform.claude.com/docs/en/about-claude/model-deprecations" }, "pricing": { "inputPerMTok": 10, "outputPerMTok": 50, "cachedInputPerMTok": 0.25, "batchDiscountPct": 50, "verifiedAt": "2026-09-08", "source": "https://platform.claude.com/docs/en/about-claude/pricing" }, "governance": { "trainsOnApiDataByDefault": false, "retentionDays": 30, "zeroDataRetentionAvailable": "no", "dataResidency": [ "global", "us" ], "hipaaBaa": true, "soc2": true, "fedramp": "none", "subprocessorsPageUrl": "https://www.anthropic.com/subprocessors", "note": "Anthropic's published certifications list for commercial services names ISO 27001:2022, ISO/IEC 42001:2023, and SOC 2 Type I and Type II. It does not name a FedRAMP authorization for the first-party Claude API. This model is a designated Covered Model and requires 30-day data retention, so zero data retention is not available for it unless Anthropic expressly authorizes it.", "verifiedAt": "2026-09-08", "sources": [ "https://www.anthropic.com/legal/commercial-terms", "https://privacy.claude.com/en/articles/7996866-how-long-do-you-store-personal-data", "https://platform.claude.com/docs/en/manage-claude/api-and-data-retention", "https://platform.claude.com/docs/en/manage-claude/data-residency", "https://privacy.claude.com/en/articles/10015870-what-certifications-has-anthropic-obtained" ] }, "availability": [ { "surface": "direct", "modelId": "claude-fable-5-1", "verifiedAt": "2026-09-08" }, { "surface": "bedrock", "modelId": "anthropic.claude-fable-5-1", "verifiedAt": "2026-09-08" }, { "surface": "vertex", "modelId": "claude-fable-5-1", "verifiedAt": "2026-09-08" }, { "surface": "azure", "modelId": "claude-fable-5-1", "note": "Microsoft Foundry deployment name.", "verifiedAt": "2026-09-08" } ] }, { "slug": "claude-opus-5", "providerId": "anthropic", "displayName": "Claude Opus 5", "apiModelIds": [ "claude-opus-5" ], "releasedOn": "2026-07-24", "family": "claude-opus", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "contextWindow": 1000000, "maxOutputTokens": 128000, "license": "proprietary", "lifecycle": { "status": "active", "sunsetNotBefore": "2027-07-24", "verifiedAt": "2026-09-08", "source": "https://platform.claude.com/docs/en/about-claude/model-deprecations" }, "pricing": { "inputPerMTok": 5, "outputPerMTok": 25, "cachedInputPerMTok": 0.5, "batchDiscountPct": 50, "verifiedAt": "2026-09-08", "source": "https://platform.claude.com/docs/en/about-claude/pricing" }, "governance": { "trainsOnApiDataByDefault": false, "retentionDays": 30, "zeroDataRetentionAvailable": "on-request", "dataResidency": [ "global", "us" ], "hipaaBaa": true, "soc2": true, "fedramp": "none", "subprocessorsPageUrl": "https://www.anthropic.com/subprocessors", "note": "Anthropic's published certifications list for commercial services names ISO 27001:2022, ISO/IEC 42001:2023, and SOC 2 Type I and Type II. It does not name a FedRAMP authorization for the first-party Claude API.", "verifiedAt": "2026-09-08", "sources": [ "https://www.anthropic.com/legal/commercial-terms", "https://privacy.claude.com/en/articles/7996866-how-long-do-you-store-personal-data", "https://platform.claude.com/docs/en/manage-claude/api-and-data-retention", "https://platform.claude.com/docs/en/manage-claude/data-residency", "https://privacy.claude.com/en/articles/10015870-what-certifications-has-anthropic-obtained" ] }, "availability": [ { "surface": "direct", "modelId": "claude-opus-5", "verifiedAt": "2026-09-08" }, { "surface": "bedrock", "modelId": "anthropic.claude-opus-5", "verifiedAt": "2026-09-08" }, { "surface": "vertex", "modelId": "claude-opus-5", "verifiedAt": "2026-09-08" }, { "surface": "azure", "modelId": "claude-opus-5", "note": "Microsoft Foundry deployment name.", "verifiedAt": "2026-09-08" } ] }, { "slug": "claude-sonnet-5", "providerId": "anthropic", "displayName": "Claude Sonnet 5", "apiModelIds": [ "claude-sonnet-5" ], "releasedOn": "2026-06-30", "family": "claude-sonnet", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "contextWindow": 1000000, "maxOutputTokens": 128000, "license": "proprietary", "lifecycle": { "status": "active", "sunsetNotBefore": "2027-06-30", "verifiedAt": "2026-09-08", "source": "https://platform.claude.com/docs/en/about-claude/model-deprecations" }, "pricing": { "inputPerMTok": 2, "outputPerMTok": 10, "cachedInputPerMTok": 0.2, "batchDiscountPct": 50, "verifiedAt": "2026-09-08", "source": "https://platform.claude.com/docs/en/about-claude/pricing" }, "governance": { "trainsOnApiDataByDefault": false, "retentionDays": 30, "zeroDataRetentionAvailable": "on-request", "dataResidency": [ "global", "us" ], "hipaaBaa": true, "soc2": true, "fedramp": "none", "subprocessorsPageUrl": "https://www.anthropic.com/subprocessors", "note": "Anthropic's published certifications list for commercial services names ISO 27001:2022, ISO/IEC 42001:2023, and SOC 2 Type I and Type II. It does not name a FedRAMP authorization for the first-party Claude API.", "verifiedAt": "2026-09-08", "sources": [ "https://www.anthropic.com/legal/commercial-terms", "https://privacy.claude.com/en/articles/7996866-how-long-do-you-store-personal-data", "https://platform.claude.com/docs/en/manage-claude/api-and-data-retention", "https://platform.claude.com/docs/en/manage-claude/data-residency", "https://privacy.claude.com/en/articles/10015870-what-certifications-has-anthropic-obtained" ] }, "availability": [ { "surface": "direct", "modelId": "claude-sonnet-5", "verifiedAt": "2026-09-08" }, { "surface": "bedrock", "modelId": "anthropic.claude-sonnet-5", "verifiedAt": "2026-09-08" }, { "surface": "vertex", "modelId": "claude-sonnet-5", "verifiedAt": "2026-09-08" }, { "surface": "azure", "modelId": "claude-sonnet-5", "note": "Microsoft Foundry deployment name.", "verifiedAt": "2026-09-08" } ] }, { "slug": "claude-haiku-4-5", "providerId": "anthropic", "displayName": "Claude Haiku 4.5", "apiModelIds": [ "claude-haiku-4-5-20251001", "claude-haiku-4-5" ], "releasedOn": "2025-10-15", "family": "claude-haiku", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "contextWindow": 200000, "maxOutputTokens": 64000, "license": "proprietary", "lifecycle": { "status": "active", "sunsetNotBefore": "2026-10-15", "verifiedAt": "2026-09-08", "source": "https://platform.claude.com/docs/en/about-claude/model-deprecations" }, "pricing": { "inputPerMTok": 1, "outputPerMTok": 5, "cachedInputPerMTok": 0.1, "batchDiscountPct": 50, "verifiedAt": "2026-09-08", "source": "https://platform.claude.com/docs/en/about-claude/pricing" }, "governance": { "trainsOnApiDataByDefault": false, "retentionDays": 30, "zeroDataRetentionAvailable": "on-request", "dataResidency": [ "global" ], "hipaaBaa": true, "soc2": true, "fedramp": "none", "subprocessorsPageUrl": "https://www.anthropic.com/subprocessors", "note": "Anthropic's published certifications list for commercial services names ISO 27001:2022, ISO/IEC 42001:2023, and SOC 2 Type I and Type II. It does not name a FedRAMP authorization for the first-party Claude API. The inference_geo parameter that pins inference to US infrastructure is supported on Claude 4.6 and later models only; requests that set it on this model return a 400 error.", "verifiedAt": "2026-09-08", "sources": [ "https://www.anthropic.com/legal/commercial-terms", "https://privacy.claude.com/en/articles/7996866-how-long-do-you-store-personal-data", "https://platform.claude.com/docs/en/manage-claude/api-and-data-retention", "https://platform.claude.com/docs/en/manage-claude/data-residency", "https://privacy.claude.com/en/articles/10015870-what-certifications-has-anthropic-obtained" ] }, "availability": [ { "surface": "direct", "modelId": "claude-haiku-4-5-20251001", "verifiedAt": "2026-09-08" }, { "surface": "bedrock", "modelId": "anthropic.claude-haiku-4-5", "verifiedAt": "2026-09-08" }, { "surface": "vertex", "modelId": "claude-haiku-4-5@20251001", "verifiedAt": "2026-09-08" }, { "surface": "azure", "modelId": "claude-haiku-4-5", "note": "Microsoft Foundry deployment name.", "verifiedAt": "2026-09-08" } ] }, { "slug": "claude-fable-5", "providerId": "anthropic", "displayName": "Claude Fable 5", "apiModelIds": [ "claude-fable-5" ], "releasedOn": "2026-06-09", "family": "claude-fable", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "contextWindow": 1000000, "maxOutputTokens": 128000, "license": "proprietary", "lifecycle": { "status": "legacy", "sunsetNotBefore": "2027-06-09", "verifiedAt": "2026-09-08", "source": "https://platform.claude.com/docs/en/about-claude/model-deprecations" }, "pricing": { "inputPerMTok": 10, "outputPerMTok": 50, "cachedInputPerMTok": 1, "batchDiscountPct": 50, "verifiedAt": "2026-09-08", "source": "https://platform.claude.com/docs/en/about-claude/pricing" }, "governance": { "trainsOnApiDataByDefault": false, "retentionDays": 30, "zeroDataRetentionAvailable": "no", "dataResidency": [ "global", "us" ], "hipaaBaa": true, "soc2": true, "fedramp": "none", "subprocessorsPageUrl": "https://www.anthropic.com/subprocessors", "note": "Anthropic's published certifications list for commercial services names ISO 27001:2022, ISO/IEC 42001:2023, and SOC 2 Type I and Type II. It does not name a FedRAMP authorization for the first-party Claude API. This model is a designated Covered Model and requires 30-day data retention, so zero data retention is not available for it unless Anthropic expressly authorizes it.", "verifiedAt": "2026-09-08", "sources": [ "https://www.anthropic.com/legal/commercial-terms", "https://privacy.claude.com/en/articles/7996866-how-long-do-you-store-personal-data", "https://platform.claude.com/docs/en/manage-claude/api-and-data-retention", "https://platform.claude.com/docs/en/manage-claude/data-residency", "https://privacy.claude.com/en/articles/10015870-what-certifications-has-anthropic-obtained" ] }, "availability": [ { "surface": "direct", "modelId": "claude-fable-5", "verifiedAt": "2026-09-08" }, { "surface": "bedrock", "modelId": "anthropic.claude-fable-5", "verifiedAt": "2026-09-08" }, { "surface": "vertex", "modelId": "claude-fable-5", "verifiedAt": "2026-09-08" }, { "surface": "azure", "modelId": "claude-fable-5", "note": "Microsoft Foundry deployment name.", "verifiedAt": "2026-09-08" } ] }, { "slug": "claude-opus-4-8", "providerId": "anthropic", "displayName": "Claude Opus 4.8", "apiModelIds": [ "claude-opus-4-8" ], "releasedOn": "2026-05-28", "family": "claude-opus", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "contextWindow": 1000000, "maxOutputTokens": 128000, "license": "proprietary", "lifecycle": { "status": "legacy", "sunsetNotBefore": "2027-05-28", "verifiedAt": "2026-09-08", "source": "https://platform.claude.com/docs/en/about-claude/model-deprecations" }, "pricing": { "inputPerMTok": 5, "outputPerMTok": 25, "cachedInputPerMTok": 0.5, "batchDiscountPct": 50, "verifiedAt": "2026-09-08", "source": "https://platform.claude.com/docs/en/about-claude/pricing" }, "governance": { "trainsOnApiDataByDefault": false, "retentionDays": 30, "zeroDataRetentionAvailable": "on-request", "dataResidency": [ "global", "us" ], "hipaaBaa": true, "soc2": true, "fedramp": "none", "subprocessorsPageUrl": "https://www.anthropic.com/subprocessors", "note": "Anthropic's published certifications list for commercial services names ISO 27001:2022, ISO/IEC 42001:2023, and SOC 2 Type I and Type II. It does not name a FedRAMP authorization for the first-party Claude API.", "verifiedAt": "2026-09-08", "sources": [ "https://www.anthropic.com/legal/commercial-terms", "https://privacy.claude.com/en/articles/7996866-how-long-do-you-store-personal-data", "https://platform.claude.com/docs/en/manage-claude/api-and-data-retention", "https://platform.claude.com/docs/en/manage-claude/data-residency", "https://privacy.claude.com/en/articles/10015870-what-certifications-has-anthropic-obtained" ] }, "availability": [ { "surface": "direct", "modelId": "claude-opus-4-8", "verifiedAt": "2026-09-08" }, { "surface": "bedrock", "modelId": "anthropic.claude-opus-4-8", "verifiedAt": "2026-09-08" }, { "surface": "vertex", "modelId": "claude-opus-4-8", "verifiedAt": "2026-09-08" }, { "surface": "azure", "modelId": "claude-opus-4-8", "note": "Microsoft Foundry deployment name.", "verifiedAt": "2026-09-08" } ] }, { "slug": "claude-opus-4-7", "providerId": "anthropic", "displayName": "Claude Opus 4.7", "apiModelIds": [ "claude-opus-4-7" ], "releasedOn": "2026-04-16", "family": "claude-opus", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "contextWindow": 1000000, "maxOutputTokens": 128000, "license": "proprietary", "lifecycle": { "status": "legacy", "sunsetNotBefore": "2027-04-16", "verifiedAt": "2026-09-08", "source": "https://platform.claude.com/docs/en/about-claude/model-deprecations" }, "pricing": { "inputPerMTok": 5, "outputPerMTok": 25, "cachedInputPerMTok": 0.5, "batchDiscountPct": 50, "verifiedAt": "2026-09-08", "source": "https://platform.claude.com/docs/en/about-claude/pricing" }, "governance": { "trainsOnApiDataByDefault": false, "retentionDays": 30, "zeroDataRetentionAvailable": "on-request", "dataResidency": [ "global", "us" ], "hipaaBaa": true, "soc2": true, "fedramp": "none", "subprocessorsPageUrl": "https://www.anthropic.com/subprocessors", "note": "Anthropic's published certifications list for commercial services names ISO 27001:2022, ISO/IEC 42001:2023, and SOC 2 Type I and Type II. It does not name a FedRAMP authorization for the first-party Claude API.", "verifiedAt": "2026-09-08", "sources": [ "https://www.anthropic.com/legal/commercial-terms", "https://privacy.claude.com/en/articles/7996866-how-long-do-you-store-personal-data", "https://platform.claude.com/docs/en/manage-claude/api-and-data-retention", "https://platform.claude.com/docs/en/manage-claude/data-residency", "https://privacy.claude.com/en/articles/10015870-what-certifications-has-anthropic-obtained" ] }, "availability": [ { "surface": "direct", "modelId": "claude-opus-4-7", "verifiedAt": "2026-09-08" }, { "surface": "bedrock", "modelId": "anthropic.claude-opus-4-7", "verifiedAt": "2026-09-08" }, { "surface": "vertex", "modelId": "claude-opus-4-7", "verifiedAt": "2026-09-08" }, { "surface": "azure", "modelId": "claude-opus-4-7", "note": "Microsoft Foundry deployment name.", "verifiedAt": "2026-09-08" } ] }, { "slug": "claude-opus-4-6", "providerId": "anthropic", "displayName": "Claude Opus 4.6", "apiModelIds": [ "claude-opus-4-6" ], "releasedOn": "2026-02-05", "family": "claude-opus", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "contextWindow": 1000000, "maxOutputTokens": 128000, "license": "proprietary", "lifecycle": { "status": "legacy", "sunsetNotBefore": "2027-02-05", "verifiedAt": "2026-09-08", "source": "https://platform.claude.com/docs/en/about-claude/model-deprecations" }, "pricing": { "inputPerMTok": 5, "outputPerMTok": 25, "cachedInputPerMTok": 0.5, "batchDiscountPct": 50, "verifiedAt": "2026-09-08", "source": "https://platform.claude.com/docs/en/about-claude/pricing" }, "governance": { "trainsOnApiDataByDefault": false, "retentionDays": 30, "zeroDataRetentionAvailable": "on-request", "dataResidency": [ "global", "us" ], "hipaaBaa": true, "soc2": true, "fedramp": "none", "subprocessorsPageUrl": "https://www.anthropic.com/subprocessors", "note": "Anthropic's published certifications list for commercial services names ISO 27001:2022, ISO/IEC 42001:2023, and SOC 2 Type I and Type II. It does not name a FedRAMP authorization for the first-party Claude API.", "verifiedAt": "2026-09-08", "sources": [ "https://www.anthropic.com/legal/commercial-terms", "https://privacy.claude.com/en/articles/7996866-how-long-do-you-store-personal-data", "https://platform.claude.com/docs/en/manage-claude/api-and-data-retention", "https://platform.claude.com/docs/en/manage-claude/data-residency", "https://privacy.claude.com/en/articles/10015870-what-certifications-has-anthropic-obtained" ] }, "availability": [ { "surface": "direct", "modelId": "claude-opus-4-6", "verifiedAt": "2026-09-08" }, { "surface": "bedrock", "modelId": "anthropic.claude-opus-4-6-v1", "note": "Bedrock InvokeModel integration.", "verifiedAt": "2026-09-08" }, { "surface": "vertex", "modelId": "claude-opus-4-6", "verifiedAt": "2026-09-08" }, { "surface": "azure", "modelId": "claude-opus-4-6", "note": "Microsoft Foundry deployment name.", "verifiedAt": "2026-09-08" } ] }, { "slug": "claude-opus-4-5", "providerId": "anthropic", "displayName": "Claude Opus 4.5", "apiModelIds": [ "claude-opus-4-5-20251101", "claude-opus-4-5" ], "releasedOn": "2025-11-24", "family": "claude-opus", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "contextWindow": 200000, "maxOutputTokens": 64000, "license": "proprietary", "lifecycle": { "status": "legacy", "sunsetNotBefore": "2026-11-24", "verifiedAt": "2026-09-08", "source": "https://platform.claude.com/docs/en/about-claude/model-deprecations" }, "pricing": { "inputPerMTok": 5, "outputPerMTok": 25, "cachedInputPerMTok": 0.5, "batchDiscountPct": 50, "verifiedAt": "2026-09-08", "source": "https://platform.claude.com/docs/en/about-claude/pricing" }, "governance": { "trainsOnApiDataByDefault": false, "retentionDays": 30, "zeroDataRetentionAvailable": "on-request", "dataResidency": [ "global" ], "hipaaBaa": true, "soc2": true, "fedramp": "none", "subprocessorsPageUrl": "https://www.anthropic.com/subprocessors", "note": "Anthropic's published certifications list for commercial services names ISO 27001:2022, ISO/IEC 42001:2023, and SOC 2 Type I and Type II. It does not name a FedRAMP authorization for the first-party Claude API. The inference_geo parameter that pins inference to US infrastructure is supported on Claude 4.6 and later models only; requests that set it on this model return a 400 error.", "verifiedAt": "2026-09-08", "sources": [ "https://www.anthropic.com/legal/commercial-terms", "https://privacy.claude.com/en/articles/7996866-how-long-do-you-store-personal-data", "https://platform.claude.com/docs/en/manage-claude/api-and-data-retention", "https://platform.claude.com/docs/en/manage-claude/data-residency", "https://privacy.claude.com/en/articles/10015870-what-certifications-has-anthropic-obtained" ] }, "availability": [ { "surface": "direct", "modelId": "claude-opus-4-5-20251101", "verifiedAt": "2026-09-08" }, { "surface": "bedrock", "modelId": "anthropic.claude-opus-4-5-20251101-v1:0", "note": "Bedrock InvokeModel integration.", "verifiedAt": "2026-09-08" }, { "surface": "vertex", "modelId": "claude-opus-4-5@20251101", "verifiedAt": "2026-09-08" }, { "surface": "azure", "modelId": "claude-opus-4-5", "note": "Microsoft Foundry deployment name.", "verifiedAt": "2026-09-08" } ] }, { "slug": "claude-sonnet-4-6", "providerId": "anthropic", "displayName": "Claude Sonnet 4.6", "apiModelIds": [ "claude-sonnet-4-6" ], "releasedOn": "2026-02-17", "family": "claude-sonnet", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "contextWindow": 1000000, "maxOutputTokens": 128000, "license": "proprietary", "lifecycle": { "status": "legacy", "sunsetNotBefore": "2027-02-17", "verifiedAt": "2026-09-08", "source": "https://platform.claude.com/docs/en/about-claude/model-deprecations" }, "pricing": { "inputPerMTok": 3, "outputPerMTok": 15, "cachedInputPerMTok": 0.3, "batchDiscountPct": 50, "verifiedAt": "2026-09-08", "source": "https://platform.claude.com/docs/en/about-claude/pricing" }, "governance": { "trainsOnApiDataByDefault": false, "retentionDays": 30, "zeroDataRetentionAvailable": "on-request", "dataResidency": [ "global", "us" ], "hipaaBaa": true, "soc2": true, "fedramp": "none", "subprocessorsPageUrl": "https://www.anthropic.com/subprocessors", "note": "Anthropic's published certifications list for commercial services names ISO 27001:2022, ISO/IEC 42001:2023, and SOC 2 Type I and Type II. It does not name a FedRAMP authorization for the first-party Claude API.", "verifiedAt": "2026-09-08", "sources": [ "https://www.anthropic.com/legal/commercial-terms", "https://privacy.claude.com/en/articles/7996866-how-long-do-you-store-personal-data", "https://platform.claude.com/docs/en/manage-claude/api-and-data-retention", "https://platform.claude.com/docs/en/manage-claude/data-residency", "https://privacy.claude.com/en/articles/10015870-what-certifications-has-anthropic-obtained" ] }, "availability": [ { "surface": "direct", "modelId": "claude-sonnet-4-6", "verifiedAt": "2026-09-08" }, { "surface": "bedrock", "modelId": "anthropic.claude-sonnet-4-6", "note": "Bedrock InvokeModel integration.", "verifiedAt": "2026-09-08" }, { "surface": "vertex", "modelId": "claude-sonnet-4-6", "verifiedAt": "2026-09-08" }, { "surface": "azure", "modelId": "claude-sonnet-4-6", "note": "Microsoft Foundry deployment name.", "verifiedAt": "2026-09-08" } ] }, { "slug": "claude-sonnet-4-5", "providerId": "anthropic", "displayName": "Claude Sonnet 4.5", "apiModelIds": [ "claude-sonnet-4-5-20250929", "claude-sonnet-4-5" ], "releasedOn": "2025-09-29", "family": "claude-sonnet", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "contextWindow": 200000, "maxOutputTokens": 64000, "license": "proprietary", "lifecycle": { "status": "legacy", "sunsetNotBefore": "2026-09-29", "verifiedAt": "2026-09-08", "source": "https://platform.claude.com/docs/en/about-claude/model-deprecations" }, "pricing": { "inputPerMTok": 3, "outputPerMTok": 15, "cachedInputPerMTok": 0.3, "batchDiscountPct": 50, "verifiedAt": "2026-09-08", "source": "https://platform.claude.com/docs/en/about-claude/pricing" }, "governance": { "trainsOnApiDataByDefault": false, "retentionDays": 30, "zeroDataRetentionAvailable": "on-request", "dataResidency": [ "global" ], "hipaaBaa": true, "soc2": true, "fedramp": "none", "subprocessorsPageUrl": "https://www.anthropic.com/subprocessors", "note": "Anthropic's published certifications list for commercial services names ISO 27001:2022, ISO/IEC 42001:2023, and SOC 2 Type I and Type II. It does not name a FedRAMP authorization for the first-party Claude API. The inference_geo parameter that pins inference to US infrastructure is supported on Claude 4.6 and later models only; requests that set it on this model return a 400 error.", "verifiedAt": "2026-09-08", "sources": [ "https://www.anthropic.com/legal/commercial-terms", "https://privacy.claude.com/en/articles/7996866-how-long-do-you-store-personal-data", "https://platform.claude.com/docs/en/manage-claude/api-and-data-retention", "https://platform.claude.com/docs/en/manage-claude/data-residency", "https://privacy.claude.com/en/articles/10015870-what-certifications-has-anthropic-obtained" ] }, "availability": [ { "surface": "direct", "modelId": "claude-sonnet-4-5-20250929", "verifiedAt": "2026-09-08" }, { "surface": "bedrock", "modelId": "anthropic.claude-sonnet-4-5-20250929-v1:0", "note": "Bedrock InvokeModel integration.", "verifiedAt": "2026-09-08" }, { "surface": "vertex", "modelId": "claude-sonnet-4-5@20250929", "verifiedAt": "2026-09-08" }, { "surface": "azure", "modelId": "claude-sonnet-4-5", "note": "Microsoft Foundry deployment name.", "verifiedAt": "2026-09-08" } ] }, { "slug": "claude-opus-4-1", "providerId": "anthropic", "displayName": "Claude Opus 4.1", "apiModelIds": [ "claude-opus-4-1-20250805" ], "family": "claude-opus", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "license": "proprietary", "lifecycle": { "status": "retired", "sunsetOn": "2026-08-05", "replacedBy": "claude-opus-4-8", "verifiedAt": "2026-09-08", "source": "https://platform.claude.com/docs/en/about-claude/model-deprecations" }, "pricing": { "inputPerMTok": 15, "outputPerMTok": 75, "cachedInputPerMTok": 1.5, "batchDiscountPct": 50, "verifiedAt": "2026-09-08", "source": "https://platform.claude.com/docs/en/about-claude/pricing" }, "governance": { "trainsOnApiDataByDefault": false, "retentionDays": 30, "zeroDataRetentionAvailable": "on-request", "dataResidency": [ "global" ], "hipaaBaa": true, "soc2": true, "fedramp": "none", "subprocessorsPageUrl": "https://www.anthropic.com/subprocessors", "note": "Anthropic's published certifications list for commercial services names ISO 27001:2022, ISO/IEC 42001:2023, and SOC 2 Type I and Type II. It does not name a FedRAMP authorization for the first-party Claude API. The inference_geo parameter that pins inference to US infrastructure is supported on Claude 4.6 and later models only; requests that set it on this model return a 400 error.", "verifiedAt": "2026-09-08", "sources": [ "https://www.anthropic.com/legal/commercial-terms", "https://privacy.claude.com/en/articles/7996866-how-long-do-you-store-personal-data", "https://platform.claude.com/docs/en/manage-claude/api-and-data-retention", "https://platform.claude.com/docs/en/manage-claude/data-residency", "https://privacy.claude.com/en/articles/10015870-what-certifications-has-anthropic-obtained" ] }, "availability": [ { "surface": "direct", "note": "Retired from the first-party Claude API; requests return an error.", "verifiedAt": "2026-09-08" }, { "surface": "bedrock", "note": "Still available on Amazon Bedrock; retirement schedule is set by AWS.", "verifiedAt": "2026-09-08" }, { "surface": "vertex", "note": "Still available on Google Cloud; retirement schedule is set by Google Cloud.", "verifiedAt": "2026-09-08" } ], "benchmarksNote": "This model's per-model documentation page was withdrawn on retirement, so its context window, max output tokens, and release date have no first-party source and are not published here." }, { "slug": "claude-opus-4", "providerId": "anthropic", "displayName": "Claude Opus 4", "apiModelIds": [ "claude-opus-4-20250514" ], "family": "claude-opus", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "license": "proprietary", "lifecycle": { "status": "retired", "sunsetOn": "2026-06-15", "replacedBy": "claude-opus-4-8", "verifiedAt": "2026-09-08", "source": "https://platform.claude.com/docs/en/about-claude/model-deprecations" }, "pricing": { "inputPerMTok": 15, "outputPerMTok": 75, "cachedInputPerMTok": 1.5, "batchDiscountPct": 50, "verifiedAt": "2026-09-08", "source": "https://platform.claude.com/docs/en/about-claude/pricing" }, "governance": { "trainsOnApiDataByDefault": false, "retentionDays": 30, "zeroDataRetentionAvailable": "on-request", "dataResidency": [ "global" ], "hipaaBaa": true, "soc2": true, "fedramp": "none", "subprocessorsPageUrl": "https://www.anthropic.com/subprocessors", "note": "Anthropic's published certifications list for commercial services names ISO 27001:2022, ISO/IEC 42001:2023, and SOC 2 Type I and Type II. It does not name a FedRAMP authorization for the first-party Claude API. The inference_geo parameter that pins inference to US infrastructure is supported on Claude 4.6 and later models only; requests that set it on this model return a 400 error.", "verifiedAt": "2026-09-08", "sources": [ "https://www.anthropic.com/legal/commercial-terms", "https://privacy.claude.com/en/articles/7996866-how-long-do-you-store-personal-data", "https://platform.claude.com/docs/en/manage-claude/api-and-data-retention", "https://platform.claude.com/docs/en/manage-claude/data-residency", "https://privacy.claude.com/en/articles/10015870-what-certifications-has-anthropic-obtained" ] }, "availability": [ { "surface": "direct", "note": "Retired from the first-party Claude API; requests return an error.", "verifiedAt": "2026-09-08" }, { "surface": "bedrock", "note": "Still available on Amazon Bedrock; retirement schedule is set by AWS.", "verifiedAt": "2026-09-08" }, { "surface": "vertex", "note": "Still available on Google Cloud; retirement schedule is set by Google Cloud.", "verifiedAt": "2026-09-08" } ], "benchmarksNote": "This model's per-model documentation page was withdrawn on retirement, so its context window, max output tokens, and release date have no first-party source and are not published here." }, { "slug": "claude-sonnet-4", "providerId": "anthropic", "displayName": "Claude Sonnet 4", "apiModelIds": [ "claude-sonnet-4-20250514" ], "family": "claude-sonnet", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "license": "proprietary", "lifecycle": { "status": "retired", "sunsetOn": "2026-06-15", "replacedBy": "claude-sonnet-4-6", "verifiedAt": "2026-09-08", "source": "https://platform.claude.com/docs/en/about-claude/model-deprecations" }, "pricing": { "inputPerMTok": 3, "outputPerMTok": 15, "cachedInputPerMTok": 0.3, "batchDiscountPct": 50, "verifiedAt": "2026-09-08", "source": "https://platform.claude.com/docs/en/about-claude/pricing" }, "governance": { "trainsOnApiDataByDefault": false, "retentionDays": 30, "zeroDataRetentionAvailable": "on-request", "dataResidency": [ "global" ], "hipaaBaa": true, "soc2": true, "fedramp": "none", "subprocessorsPageUrl": "https://www.anthropic.com/subprocessors", "note": "Anthropic's published certifications list for commercial services names ISO 27001:2022, ISO/IEC 42001:2023, and SOC 2 Type I and Type II. It does not name a FedRAMP authorization for the first-party Claude API. The inference_geo parameter that pins inference to US infrastructure is supported on Claude 4.6 and later models only; requests that set it on this model return a 400 error.", "verifiedAt": "2026-09-08", "sources": [ "https://www.anthropic.com/legal/commercial-terms", "https://privacy.claude.com/en/articles/7996866-how-long-do-you-store-personal-data", "https://platform.claude.com/docs/en/manage-claude/api-and-data-retention", "https://platform.claude.com/docs/en/manage-claude/data-residency", "https://privacy.claude.com/en/articles/10015870-what-certifications-has-anthropic-obtained" ] }, "availability": [ { "surface": "direct", "note": "Retired from the first-party Claude API; requests return an error.", "verifiedAt": "2026-09-08" }, { "surface": "bedrock", "note": "Still available on Amazon Bedrock; retirement schedule is set by AWS.", "verifiedAt": "2026-09-08" }, { "surface": "vertex", "note": "Still available on Google Cloud; retirement schedule is set by Google Cloud.", "verifiedAt": "2026-09-08" } ], "benchmarksNote": "This model's per-model documentation page was withdrawn on retirement, so its context window, max output tokens, and release date have no first-party source and are not published here." }, { "slug": "claude-sonnet-3-7", "providerId": "anthropic", "displayName": "Claude Sonnet 3.7", "apiModelIds": [ "claude-3-7-sonnet-20250219" ], "family": "claude-sonnet", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "license": "proprietary", "lifecycle": { "status": "retired", "sunsetOn": "2026-02-19", "replacedBy": "claude-sonnet-4-6", "verifiedAt": "2026-09-08", "source": "https://platform.claude.com/docs/en/about-claude/model-deprecations" }, "pricing": { "inputPerMTok": 3, "outputPerMTok": 15, "verifiedAt": "2026-09-08", "source": "https://www.anthropic.com/news/claude-3-7-sonnet" }, "governance": { "trainsOnApiDataByDefault": false, "retentionDays": 30, "zeroDataRetentionAvailable": "on-request", "dataResidency": [ "global" ], "hipaaBaa": true, "soc2": true, "fedramp": "none", "subprocessorsPageUrl": "https://www.anthropic.com/subprocessors", "note": "Anthropic's published certifications list for commercial services names ISO 27001:2022, ISO/IEC 42001:2023, and SOC 2 Type I and Type II. It does not name a FedRAMP authorization for the first-party Claude API. The inference_geo parameter that pins inference to US infrastructure is supported on Claude 4.6 and later models only; requests that set it on this model return a 400 error.", "verifiedAt": "2026-09-08", "sources": [ "https://www.anthropic.com/legal/commercial-terms", "https://privacy.claude.com/en/articles/7996866-how-long-do-you-store-personal-data", "https://platform.claude.com/docs/en/manage-claude/api-and-data-retention", "https://platform.claude.com/docs/en/manage-claude/data-residency", "https://privacy.claude.com/en/articles/10015870-what-certifications-has-anthropic-obtained" ] }, "availability": [ { "surface": "direct", "note": "Retired from the first-party Claude API; requests return an error.", "verifiedAt": "2026-09-08" } ], "benchmarksNote": "This model's per-model documentation page was withdrawn on retirement, so its context window, max output tokens, and release date have no first-party source and are not published here." }, { "slug": "claude-haiku-3-5", "providerId": "anthropic", "displayName": "Claude Haiku 3.5", "apiModelIds": [ "claude-3-5-haiku-20241022" ], "family": "claude-haiku", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "license": "proprietary", "lifecycle": { "status": "retired", "sunsetOn": "2026-02-19", "replacedBy": "claude-haiku-4-5", "verifiedAt": "2026-09-08", "source": "https://platform.claude.com/docs/en/about-claude/model-deprecations" }, "pricing": { "inputPerMTok": 0.8, "outputPerMTok": 4, "cachedInputPerMTok": 0.08, "batchDiscountPct": 50, "verifiedAt": "2026-09-08", "source": "https://platform.claude.com/docs/en/about-claude/pricing" }, "governance": { "trainsOnApiDataByDefault": false, "retentionDays": 30, "zeroDataRetentionAvailable": "on-request", "dataResidency": [ "global" ], "hipaaBaa": true, "soc2": true, "fedramp": "none", "subprocessorsPageUrl": "https://www.anthropic.com/subprocessors", "note": "Anthropic's published certifications list for commercial services names ISO 27001:2022, ISO/IEC 42001:2023, and SOC 2 Type I and Type II. It does not name a FedRAMP authorization for the first-party Claude API. The inference_geo parameter that pins inference to US infrastructure is supported on Claude 4.6 and later models only; requests that set it on this model return a 400 error.", "verifiedAt": "2026-09-08", "sources": [ "https://www.anthropic.com/legal/commercial-terms", "https://privacy.claude.com/en/articles/7996866-how-long-do-you-store-personal-data", "https://platform.claude.com/docs/en/manage-claude/api-and-data-retention", "https://platform.claude.com/docs/en/manage-claude/data-residency", "https://privacy.claude.com/en/articles/10015870-what-certifications-has-anthropic-obtained" ] }, "availability": [ { "surface": "direct", "note": "Retired from the first-party Claude API; requests return an error.", "verifiedAt": "2026-09-08" }, { "surface": "bedrock", "note": "Still available on Amazon Bedrock; retirement schedule is set by AWS.", "verifiedAt": "2026-09-08" }, { "surface": "vertex", "note": "Still available on Google Cloud; retirement schedule is set by Google Cloud.", "verifiedAt": "2026-09-08" } ], "benchmarksNote": "This model's per-model documentation page was withdrawn on retirement, so its context window, max output tokens, and release date have no first-party source and are not published here." }, { "slug": "claude-haiku-3", "providerId": "anthropic", "displayName": "Claude Haiku 3", "apiModelIds": [ "claude-3-haiku-20240307" ], "family": "claude-haiku", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "license": "proprietary", "lifecycle": { "status": "retired", "sunsetOn": "2026-04-20", "replacedBy": "claude-haiku-4-5", "verifiedAt": "2026-09-08", "source": "https://platform.claude.com/docs/en/about-claude/model-deprecations" }, "pricing": { "inputPerMTok": 0.25, "outputPerMTok": 1.25, "verifiedAt": "2026-09-08", "source": "https://www.anthropic.com/news/claude-3-family" }, "governance": { "trainsOnApiDataByDefault": false, "retentionDays": 30, "zeroDataRetentionAvailable": "on-request", "dataResidency": [ "global" ], "hipaaBaa": true, "soc2": true, "fedramp": "none", "subprocessorsPageUrl": "https://www.anthropic.com/subprocessors", "note": "Anthropic's published certifications list for commercial services names ISO 27001:2022, ISO/IEC 42001:2023, and SOC 2 Type I and Type II. It does not name a FedRAMP authorization for the first-party Claude API. The inference_geo parameter that pins inference to US infrastructure is supported on Claude 4.6 and later models only; requests that set it on this model return a 400 error.", "verifiedAt": "2026-09-08", "sources": [ "https://www.anthropic.com/legal/commercial-terms", "https://privacy.claude.com/en/articles/7996866-how-long-do-you-store-personal-data", "https://platform.claude.com/docs/en/manage-claude/api-and-data-retention", "https://platform.claude.com/docs/en/manage-claude/data-residency", "https://privacy.claude.com/en/articles/10015870-what-certifications-has-anthropic-obtained" ] }, "availability": [ { "surface": "direct", "note": "Retired from the first-party Claude API; requests return an error.", "verifiedAt": "2026-09-08" } ], "benchmarksNote": "This model's per-model documentation page was withdrawn on retirement, so its context window, max output tokens, and release date have no first-party source and are not published here." }, { "slug": "gpt-6-astra", "providerId": "openai", "displayName": "GPT-6 Astra", "apiModelIds": [ "gpt-6-astra" ], "releasedOn": "2026-09-03", "family": "gpt-6", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "contextWindow": 1050000, "maxOutputTokens": 128000, "license": "proprietary", "lifecycle": { "status": "active", "verifiedAt": "2026-09-08", "source": "https://developers.openai.com/api/docs/deprecations" }, "pricing": { "inputPerMTok": 10, "outputPerMTok": 50, "cachedInputPerMTok": 1, "batchDiscountPct": 50, "verifiedAt": "2026-09-08", "source": "https://developers.openai.com/api/docs/pricing" }, "governance": { "trainsOnApiDataByDefault": false, "retentionDays": 30, "zeroDataRetentionAvailable": "on-request", "dataResidency": [ "us", "eu", "gb", "ca", "au", "jp", "in", "sg", "kr", "ae" ], "hipaaBaa": true, "soc2": true, "fedramp": "unknown", "note": "Data sent to the OpenAI API is not used to train or improve OpenAI models unless a customer explicitly opts in. The 30 day figure is the abuse monitoring log retention window. Zero data retention is available to eligible customers subject to prior approval by OpenAI. The Trust Portal lists SOC 2 Type 2 and FedRAMP 20x documentation but does not publish a FedRAMP impact level, and a Business Associate Addendum is available for eligible endpoints used with zero data retention or modified abuse monitoring.", "verifiedAt": "2026-09-08", "sources": [ "https://developers.openai.com/api/docs/guides/your-data", "https://trust.openai.com/" ] }, "availability": [ { "surface": "direct", "modelId": "gpt-6-astra", "note": "First-party OpenAI API.", "verifiedAt": "2026-09-08" } ], "benchmarksNote": "OpenAI prices prompts above 272,000 input tokens at 2x the standard input and cache rates and 1.5x the standard output rate for the full request on this model." }, { "slug": "gpt-5-6-sol", "providerId": "openai", "displayName": "GPT-5.6 Sol", "apiModelIds": [ "gpt-5.6-sol" ], "releasedOn": "2026-07-09", "family": "gpt-5.6", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "contextWindow": 1050000, "maxOutputTokens": 128000, "license": "proprietary", "lifecycle": { "status": "active", "verifiedAt": "2026-09-08", "source": "https://developers.openai.com/api/docs/deprecations" }, "pricing": { "inputPerMTok": 4, "outputPerMTok": 20, "cachedInputPerMTok": 0.4, "batchDiscountPct": 50, "verifiedAt": "2026-09-08", "source": "https://developers.openai.com/api/docs/pricing" }, "governance": { "trainsOnApiDataByDefault": false, "retentionDays": 30, "zeroDataRetentionAvailable": "on-request", "dataResidency": [ "us", "eu", "gb", "ca", "au", "jp", "in", "sg", "kr", "ae" ], "hipaaBaa": true, "soc2": true, "fedramp": "unknown", "note": "Data sent to the OpenAI API is not used to train or improve OpenAI models unless a customer explicitly opts in. The 30 day figure is the abuse monitoring log retention window. Zero data retention is available to eligible customers subject to prior approval by OpenAI. The Trust Portal lists SOC 2 Type 2 and FedRAMP 20x documentation but does not publish a FedRAMP impact level, and a Business Associate Addendum is available for eligible endpoints used with zero data retention or modified abuse monitoring.", "verifiedAt": "2026-09-08", "sources": [ "https://developers.openai.com/api/docs/guides/your-data", "https://trust.openai.com/" ] }, "availability": [ { "surface": "direct", "modelId": "gpt-5.6-sol", "note": "First-party OpenAI API.", "verifiedAt": "2026-09-08" } ] }, { "slug": "gpt-5-6-terra", "providerId": "openai", "displayName": "GPT-5.6 Terra", "apiModelIds": [ "gpt-5.6-terra" ], "releasedOn": "2026-07-09", "family": "gpt-5.6", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "contextWindow": 1050000, "maxOutputTokens": 128000, "license": "proprietary", "lifecycle": { "status": "active", "verifiedAt": "2026-09-08", "source": "https://developers.openai.com/api/docs/deprecations" }, "pricing": { "inputPerMTok": 2, "outputPerMTok": 12, "cachedInputPerMTok": 0.2, "batchDiscountPct": 50, "verifiedAt": "2026-09-08", "source": "https://developers.openai.com/api/docs/pricing" }, "governance": { "trainsOnApiDataByDefault": false, "retentionDays": 30, "zeroDataRetentionAvailable": "on-request", "dataResidency": [ "us", "eu", "gb", "ca", "au", "jp", "in", "sg", "kr", "ae" ], "hipaaBaa": true, "soc2": true, "fedramp": "unknown", "note": "Data sent to the OpenAI API is not used to train or improve OpenAI models unless a customer explicitly opts in. The 30 day figure is the abuse monitoring log retention window. Zero data retention is available to eligible customers subject to prior approval by OpenAI. The Trust Portal lists SOC 2 Type 2 and FedRAMP 20x documentation but does not publish a FedRAMP impact level, and a Business Associate Addendum is available for eligible endpoints used with zero data retention or modified abuse monitoring.", "verifiedAt": "2026-09-08", "sources": [ "https://developers.openai.com/api/docs/guides/your-data", "https://trust.openai.com/" ] }, "availability": [ { "surface": "direct", "modelId": "gpt-5.6-terra", "note": "First-party OpenAI API.", "verifiedAt": "2026-09-08" } ] }, { "slug": "gpt-5-6-luna", "providerId": "openai", "displayName": "GPT-5.6 Luna", "apiModelIds": [ "gpt-5.6-luna" ], "releasedOn": "2026-07-09", "family": "gpt-5.6", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "contextWindow": 1050000, "maxOutputTokens": 128000, "license": "proprietary", "lifecycle": { "status": "active", "verifiedAt": "2026-09-08", "source": "https://developers.openai.com/api/docs/deprecations" }, "pricing": { "inputPerMTok": 0.2, "outputPerMTok": 1.2, "cachedInputPerMTok": 0.02, "batchDiscountPct": 50, "verifiedAt": "2026-09-08", "source": "https://developers.openai.com/api/docs/pricing" }, "governance": { "trainsOnApiDataByDefault": false, "retentionDays": 30, "zeroDataRetentionAvailable": "on-request", "dataResidency": [ "us", "eu", "gb", "ca", "au", "jp", "in", "sg", "kr", "ae" ], "hipaaBaa": true, "soc2": true, "fedramp": "unknown", "note": "Data sent to the OpenAI API is not used to train or improve OpenAI models unless a customer explicitly opts in. The 30 day figure is the abuse monitoring log retention window. Zero data retention is available to eligible customers subject to prior approval by OpenAI. The Trust Portal lists SOC 2 Type 2 and FedRAMP 20x documentation but does not publish a FedRAMP impact level, and a Business Associate Addendum is available for eligible endpoints used with zero data retention or modified abuse monitoring.", "verifiedAt": "2026-09-08", "sources": [ "https://developers.openai.com/api/docs/guides/your-data", "https://trust.openai.com/" ] }, "availability": [ { "surface": "direct", "modelId": "gpt-5.6-luna", "note": "First-party OpenAI API.", "verifiedAt": "2026-09-08" } ] }, { "slug": "gpt-5-5", "providerId": "openai", "displayName": "GPT-5.5", "apiModelIds": [ "gpt-5.5", "gpt-5.5-2026-04-23" ], "releasedOn": "2026-04-24", "family": "gpt-5.5", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "contextWindow": 1050000, "maxOutputTokens": 128000, "license": "proprietary", "lifecycle": { "status": "active", "verifiedAt": "2026-09-08", "source": "https://developers.openai.com/api/docs/deprecations" }, "pricing": { "inputPerMTok": 5, "outputPerMTok": 30, "cachedInputPerMTok": 0.5, "batchDiscountPct": 50, "verifiedAt": "2026-09-08", "source": "https://developers.openai.com/api/docs/pricing" }, "governance": { "trainsOnApiDataByDefault": false, "retentionDays": 30, "zeroDataRetentionAvailable": "on-request", "dataResidency": [ "us", "eu", "gb", "ca", "au", "jp", "in", "sg", "kr", "ae" ], "hipaaBaa": true, "soc2": true, "fedramp": "unknown", "note": "Data sent to the OpenAI API is not used to train or improve OpenAI models unless a customer explicitly opts in. The 30 day figure is the abuse monitoring log retention window. Zero data retention is available to eligible customers subject to prior approval by OpenAI. The Trust Portal lists SOC 2 Type 2 and FedRAMP 20x documentation but does not publish a FedRAMP impact level, and a Business Associate Addendum is available for eligible endpoints used with zero data retention or modified abuse monitoring.", "verifiedAt": "2026-09-08", "sources": [ "https://developers.openai.com/api/docs/guides/your-data", "https://trust.openai.com/" ] }, "availability": [ { "surface": "direct", "modelId": "gpt-5.5", "note": "First-party OpenAI API.", "verifiedAt": "2026-09-08" } ] }, { "slug": "gpt-5-4", "providerId": "openai", "displayName": "GPT-5.4", "apiModelIds": [ "gpt-5.4", "gpt-5.4-2026-03-05" ], "releasedOn": "2026-03-05", "family": "gpt-5.4", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "contextWindow": 1050000, "maxOutputTokens": 128000, "license": "proprietary", "lifecycle": { "status": "active", "verifiedAt": "2026-09-08", "source": "https://developers.openai.com/api/docs/deprecations" }, "pricing": { "inputPerMTok": 2.5, "outputPerMTok": 15, "cachedInputPerMTok": 0.25, "batchDiscountPct": 50, "verifiedAt": "2026-09-08", "source": "https://developers.openai.com/api/docs/pricing" }, "governance": { "trainsOnApiDataByDefault": false, "retentionDays": 30, "zeroDataRetentionAvailable": "on-request", "dataResidency": [ "us", "eu", "gb", "ca", "au", "jp", "in", "sg", "kr", "ae" ], "hipaaBaa": true, "soc2": true, "fedramp": "unknown", "note": "Data sent to the OpenAI API is not used to train or improve OpenAI models unless a customer explicitly opts in. The 30 day figure is the abuse monitoring log retention window. Zero data retention is available to eligible customers subject to prior approval by OpenAI. The Trust Portal lists SOC 2 Type 2 and FedRAMP 20x documentation but does not publish a FedRAMP impact level, and a Business Associate Addendum is available for eligible endpoints used with zero data retention or modified abuse monitoring.", "verifiedAt": "2026-09-08", "sources": [ "https://developers.openai.com/api/docs/guides/your-data", "https://trust.openai.com/" ] }, "availability": [ { "surface": "direct", "modelId": "gpt-5.4", "note": "First-party OpenAI API.", "verifiedAt": "2026-09-08" } ] }, { "slug": "gpt-5", "providerId": "openai", "displayName": "GPT-5", "apiModelIds": [ "gpt-5", "gpt-5-2025-08-07" ], "releasedOn": "2025-08-07", "family": "gpt-5", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "contextWindow": 400000, "maxOutputTokens": 128000, "license": "proprietary", "lifecycle": { "status": "deprecated", "sunsetOn": "2026-12-11", "replacedBy": "gpt-5-6-sol", "verifiedAt": "2026-09-08", "source": "https://developers.openai.com/api/docs/deprecations" }, "pricing": { "inputPerMTok": 1.25, "outputPerMTok": 10, "cachedInputPerMTok": 0.125, "batchDiscountPct": 50, "verifiedAt": "2026-09-08", "source": "https://developers.openai.com/api/docs/pricing" }, "governance": { "trainsOnApiDataByDefault": false, "retentionDays": 30, "zeroDataRetentionAvailable": "on-request", "dataResidency": [ "us", "eu", "gb", "ca", "au", "jp", "in", "sg", "kr", "ae" ], "hipaaBaa": true, "soc2": true, "fedramp": "unknown", "note": "Data sent to the OpenAI API is not used to train or improve OpenAI models unless a customer explicitly opts in. The 30 day figure is the abuse monitoring log retention window. Zero data retention is available to eligible customers subject to prior approval by OpenAI. The Trust Portal lists SOC 2 Type 2 and FedRAMP 20x documentation but does not publish a FedRAMP impact level, and a Business Associate Addendum is available for eligible endpoints used with zero data retention or modified abuse monitoring.", "verifiedAt": "2026-09-08", "sources": [ "https://developers.openai.com/api/docs/guides/your-data", "https://trust.openai.com/" ] }, "availability": [ { "surface": "direct", "modelId": "gpt-5-2025-08-07", "note": "First-party OpenAI API.", "verifiedAt": "2026-09-08" } ], "benchmarksNote": "OpenAI announced the deprecation of gpt-5-2025-08-07 on June 11, 2026, with a shutdown date of December 11, 2026 and gpt-5.6-sol as the named replacement." }, { "slug": "gemini-3-1-pro-preview", "providerId": "google", "displayName": "Gemini 3.1 Pro Preview", "apiModelIds": [ "gemini-3.1-pro-preview" ], "releasedOn": "2026-02-19", "family": "gemini-3", "modalities": { "input": [ "text", "image", "video", "audio" ], "output": [ "text" ] }, "contextWindow": 1048576, "maxOutputTokens": 65536, "license": "proprietary", "lifecycle": { "status": "active", "verifiedAt": "2026-09-08", "source": "https://ai.google.dev/gemini-api/docs/deprecations" }, "pricing": { "inputPerMTok": 2, "outputPerMTok": 12, "cachedInputPerMTok": 0.2, "batchDiscountPct": 50, "verifiedAt": "2026-09-08", "source": "https://ai.google.dev/gemini-api/docs/pricing" }, "governance": { "trainsOnApiDataByDefault": false, "retentionDays": "unknown", "zeroDataRetentionAvailable": "on-request", "dataResidency": [ "global" ], "hipaaBaa": false, "soc2": false, "fedramp": "unknown", "note": "The Gemini API Additional Terms of Service state that for Paid Services, Google does not use prompts or responses to improve its products, and logs them for a limited period solely to detect Prohibited Use Policy violations without stating a specific number of days. Zero data retention is available on request per project. Unpaid use, including Google AI Studio and the free quota, is used to develop Google products and may be reviewed by human annotators. HIPAA BAA, SOC 2, and FedRAMP coverage are published for Vertex AI and Gemini Enterprise, not for this first-party Developer API surface, so they are not claimed here.", "verifiedAt": "2026-09-08", "sources": [ "https://ai.google.dev/gemini-api/terms", "https://ai.google.dev/gemini-api/docs/zdr" ] }, "availability": [ { "surface": "direct", "modelId": "gemini-3.1-pro-preview", "note": "First-party Gemini Developer API (ai.google.dev / Google AI Studio).", "verifiedAt": "2026-09-08" } ], "benchmarksNote": "Prompts longer than 200,000 tokens are billed at $4 input and $18 output per million tokens instead of the standard $2 and $12. Output pricing includes thinking tokens." }, { "slug": "gemini-3-6-flash", "providerId": "google", "displayName": "Gemini 3.6 Flash", "apiModelIds": [ "gemini-3.6-flash" ], "releasedOn": "2026-07-21", "family": "gemini-3", "modalities": { "input": [ "text", "image", "video", "audio" ], "output": [ "text" ] }, "contextWindow": 1048576, "maxOutputTokens": 65536, "license": "proprietary", "lifecycle": { "status": "active", "verifiedAt": "2026-09-08", "source": "https://ai.google.dev/gemini-api/docs/deprecations" }, "pricing": { "inputPerMTok": 0.75, "outputPerMTok": 3.75, "cachedInputPerMTok": 0.075, "batchDiscountPct": 50, "verifiedAt": "2026-09-08", "source": "https://ai.google.dev/gemini-api/docs/pricing" }, "governance": { "trainsOnApiDataByDefault": false, "retentionDays": "unknown", "zeroDataRetentionAvailable": "on-request", "dataResidency": [ "global" ], "hipaaBaa": false, "soc2": false, "fedramp": "unknown", "note": "The Gemini API Additional Terms of Service state that for Paid Services, Google does not use prompts or responses to improve its products, and logs them for a limited period solely to detect Prohibited Use Policy violations without stating a specific number of days. Zero data retention is available on request per project. Unpaid use, including Google AI Studio and the free quota, is used to develop Google products and may be reviewed by human annotators. HIPAA BAA, SOC 2, and FedRAMP coverage are published for Vertex AI and Gemini Enterprise, not for this first-party Developer API surface, so they are not claimed here.", "verifiedAt": "2026-09-08", "sources": [ "https://ai.google.dev/gemini-api/terms", "https://ai.google.dev/gemini-api/docs/zdr" ] }, "availability": [ { "surface": "direct", "modelId": "gemini-3.6-flash", "note": "First-party Gemini Developer API (ai.google.dev / Google AI Studio).", "verifiedAt": "2026-09-08" } ], "benchmarksNote": "This introductory rate holds through December 31, 2026. From January 1, 2027 the published rate rises to $1.50 input and $7.50 output per million tokens, and cached input to $0.15. Output pricing includes thinking tokens." }, { "slug": "gemini-3-5-flash-lite", "providerId": "google", "displayName": "Gemini 3.5 Flash-Lite", "apiModelIds": [ "gemini-3.5-flash-lite" ], "releasedOn": "2026-07-21", "family": "gemini-3", "modalities": { "input": [ "text", "image", "video", "audio" ], "output": [ "text" ] }, "contextWindow": 1048576, "maxOutputTokens": 65536, "license": "proprietary", "lifecycle": { "status": "active", "verifiedAt": "2026-09-08", "source": "https://ai.google.dev/gemini-api/docs/deprecations" }, "pricing": { "inputPerMTok": 0.3, "outputPerMTok": 2.5, "cachedInputPerMTok": 0.03, "batchDiscountPct": 50, "verifiedAt": "2026-09-08", "source": "https://ai.google.dev/gemini-api/docs/pricing" }, "governance": { "trainsOnApiDataByDefault": false, "retentionDays": "unknown", "zeroDataRetentionAvailable": "on-request", "dataResidency": [ "global" ], "hipaaBaa": false, "soc2": false, "fedramp": "unknown", "note": "The Gemini API Additional Terms of Service state that for Paid Services, Google does not use prompts or responses to improve its products, and logs them for a limited period solely to detect Prohibited Use Policy violations without stating a specific number of days. Zero data retention is available on request per project. Unpaid use, including Google AI Studio and the free quota, is used to develop Google products and may be reviewed by human annotators. HIPAA BAA, SOC 2, and FedRAMP coverage are published for Vertex AI and Gemini Enterprise, not for this first-party Developer API surface, so they are not claimed here.", "verifiedAt": "2026-09-08", "sources": [ "https://ai.google.dev/gemini-api/terms", "https://ai.google.dev/gemini-api/docs/zdr" ] }, "availability": [ { "surface": "direct", "modelId": "gemini-3.5-flash-lite", "note": "First-party Gemini Developer API (ai.google.dev / Google AI Studio).", "verifiedAt": "2026-09-08" } ] }, { "slug": "gemini-2-5-pro", "providerId": "google", "displayName": "Gemini 2.5 Pro", "apiModelIds": [ "gemini-2.5-pro" ], "releasedOn": "2025-06-17", "family": "gemini-2.5", "modalities": { "input": [ "text", "image", "video", "audio" ], "output": [ "text" ] }, "contextWindow": 1048576, "maxOutputTokens": 65536, "license": "proprietary", "lifecycle": { "status": "legacy", "verifiedAt": "2026-09-08", "source": "https://ai.google.dev/gemini-api/docs/deprecations" }, "pricing": { "inputPerMTok": 1.25, "outputPerMTok": 10, "cachedInputPerMTok": 0.125, "batchDiscountPct": 50, "verifiedAt": "2026-09-08", "source": "https://ai.google.dev/gemini-api/docs/pricing" }, "governance": { "trainsOnApiDataByDefault": false, "retentionDays": "unknown", "zeroDataRetentionAvailable": "on-request", "dataResidency": [ "global" ], "hipaaBaa": false, "soc2": false, "fedramp": "unknown", "note": "The Gemini API Additional Terms of Service state that for Paid Services, Google does not use prompts or responses to improve its products, and logs them for a limited period solely to detect Prohibited Use Policy violations without stating a specific number of days. Zero data retention is available on request per project. Unpaid use, including Google AI Studio and the free quota, is used to develop Google products and may be reviewed by human annotators. HIPAA BAA, SOC 2, and FedRAMP coverage are published for Vertex AI and Gemini Enterprise, not for this first-party Developer API surface, so they are not claimed here.", "verifiedAt": "2026-09-08", "sources": [ "https://ai.google.dev/gemini-api/terms", "https://ai.google.dev/gemini-api/docs/zdr" ] }, "availability": [ { "surface": "direct", "modelId": "gemini-2.5-pro", "note": "First-party Gemini Developer API (ai.google.dev / Google AI Studio).", "verifiedAt": "2026-09-08" } ], "benchmarksNote": "No shutdown date is announced for this model as of the V1 audit. Prompts longer than 200,000 tokens are billed at $2.50 input and $15 output per million tokens instead of the standard $1.25 and $10. Output pricing includes thinking tokens." }, { "slug": "gemini-2-5-flash", "providerId": "google", "displayName": "Gemini 2.5 Flash", "apiModelIds": [ "gemini-2.5-flash" ], "releasedOn": "2025-06-17", "family": "gemini-2.5", "modalities": { "input": [ "text", "image", "video", "audio" ], "output": [ "text" ] }, "contextWindow": 1048576, "maxOutputTokens": 65536, "license": "proprietary", "lifecycle": { "status": "legacy", "verifiedAt": "2026-09-08", "source": "https://ai.google.dev/gemini-api/docs/deprecations" }, "pricing": { "inputPerMTok": 0.3, "outputPerMTok": 2.5, "cachedInputPerMTok": 0.03, "batchDiscountPct": 50, "verifiedAt": "2026-09-08", "source": "https://ai.google.dev/gemini-api/docs/pricing" }, "governance": { "trainsOnApiDataByDefault": false, "retentionDays": "unknown", "zeroDataRetentionAvailable": "on-request", "dataResidency": [ "global" ], "hipaaBaa": false, "soc2": false, "fedramp": "unknown", "note": "The Gemini API Additional Terms of Service state that for Paid Services, Google does not use prompts or responses to improve its products, and logs them for a limited period solely to detect Prohibited Use Policy violations without stating a specific number of days. Zero data retention is available on request per project. Unpaid use, including Google AI Studio and the free quota, is used to develop Google products and may be reviewed by human annotators. HIPAA BAA, SOC 2, and FedRAMP coverage are published for Vertex AI and Gemini Enterprise, not for this first-party Developer API surface, so they are not claimed here.", "verifiedAt": "2026-09-08", "sources": [ "https://ai.google.dev/gemini-api/terms", "https://ai.google.dev/gemini-api/docs/zdr" ] }, "availability": [ { "surface": "direct", "modelId": "gemini-2.5-flash", "note": "First-party Gemini Developer API (ai.google.dev / Google AI Studio).", "verifiedAt": "2026-09-08" } ], "benchmarksNote": "No shutdown date is announced for this model as of the V1 audit. Audio input is billed at $1 per million tokens instead of the standard $0.30 for text, image, and video. Output pricing includes thinking tokens." }, { "slug": "mistral-large-3", "providerId": "mistral", "displayName": "Mistral Large 3", "apiModelIds": [ "mistral-large-2512" ], "releasedOn": "2025-12-02", "family": "mistral-large", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "contextWindow": 256000, "maxOutputTokens": 256000, "license": "open-weights", "lifecycle": { "status": "active", "verifiedAt": "2026-09-08", "source": "https://docs.mistral.ai/models" }, "pricing": { "inputPerMTok": 0.5, "outputPerMTok": 1.5, "verifiedAt": "2026-09-08", "source": "https://mistral.ai/pricing/api" }, "governance": { "trainsOnApiDataByDefault": true, "retentionDays": "unknown", "zeroDataRetentionAvailable": "on-request", "dataResidency": [ "eu" ], "hipaaBaa": false, "soc2": true, "fedramp": "unknown", "note": "Mistral's help center states that using API calls to improve its services is on by default, with customers retaining the right to opt out at any time from the Admin panel's Privacy menu; the Vibe and API opt-out toggles are separate. Zero data retention is available on paid plans for supported stateless API calls but is not a self-service toggle: it requires a written request with a stated reason, which Mistral reviews and may approve or deny, after which it appears in Admin privacy settings. Data is hosted in the EU by default, with an explicit opt-in to a US API endpoint available. No specific retention period in days for API prompts and responses is stated on a page this audit could confirm, so it is recorded as unknown rather than the widely repeated but unverifiable 30-day figure. Mistral states it complies with SOC 2 Type II and ISO 27001/27701; a HIPAA Business Associate Agreement and any FedRAMP authorization are not confirmed on a page this audit could read, so they are recorded conservatively rather than assumed.", "verifiedAt": "2026-09-08", "sources": [ "https://docs.mistral.ai/admin/monitor-comply/zero-data-retention", "https://help.mistral.ai/en/articles/455207-can-i-opt-out-of-my-input-or-output-data-being-used-for-training", "https://help.mistral.ai/en/articles/347629-where-do-you-store-my-data-or-my-organization-s-data", "https://help.mistral.ai/en/articles/323772-do-you-have-soc2-or-iso27001-certification" ] }, "availability": [ { "surface": "direct", "modelId": "mistral-large-2512", "note": "First-party Mistral La Plateforme API.", "verifiedAt": "2026-09-08" }, { "surface": "self-host", "note": "Released under the Apache 2.0 license; weights published for self-hosting.", "verifiedAt": "2026-09-08" } ], "benchmarksNote": "Mistral's API reference (docs.mistral.ai/api) does not publish a max output token figure separate from context window: it states only that prompt tokens plus max_tokens must not exceed the model's context length, so the output cap recorded here equals the published context window." }, { "slug": "mistral-medium-3-5", "providerId": "mistral", "displayName": "Mistral Medium 3.5", "apiModelIds": [ "mistral-medium-2604" ], "releasedOn": "2026-04-28", "family": "mistral-medium", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "contextWindow": 256000, "maxOutputTokens": 256000, "license": "open-weights", "lifecycle": { "status": "active", "verifiedAt": "2026-09-08", "source": "https://docs.mistral.ai/models" }, "pricing": { "inputPerMTok": 1.5, "outputPerMTok": 7.5, "verifiedAt": "2026-09-08", "source": "https://mistral.ai/pricing/api" }, "governance": { "trainsOnApiDataByDefault": true, "retentionDays": "unknown", "zeroDataRetentionAvailable": "on-request", "dataResidency": [ "eu" ], "hipaaBaa": false, "soc2": true, "fedramp": "unknown", "note": "Mistral's help center states that using API calls to improve its services is on by default, with customers retaining the right to opt out at any time from the Admin panel's Privacy menu; the Vibe and API opt-out toggles are separate. Zero data retention is available on paid plans for supported stateless API calls but is not a self-service toggle: it requires a written request with a stated reason, which Mistral reviews and may approve or deny, after which it appears in Admin privacy settings. Data is hosted in the EU by default, with an explicit opt-in to a US API endpoint available. No specific retention period in days for API prompts and responses is stated on a page this audit could confirm, so it is recorded as unknown rather than the widely repeated but unverifiable 30-day figure. Mistral states it complies with SOC 2 Type II and ISO 27001/27701; a HIPAA Business Associate Agreement and any FedRAMP authorization are not confirmed on a page this audit could read, so they are recorded conservatively rather than assumed.", "verifiedAt": "2026-09-08", "sources": [ "https://docs.mistral.ai/admin/monitor-comply/zero-data-retention", "https://help.mistral.ai/en/articles/455207-can-i-opt-out-of-my-input-or-output-data-being-used-for-training", "https://help.mistral.ai/en/articles/347629-where-do-you-store-my-data-or-my-organization-s-data", "https://help.mistral.ai/en/articles/323772-do-you-have-soc2-or-iso27001-certification" ] }, "availability": [ { "surface": "direct", "modelId": "mistral-medium-2604", "note": "First-party Mistral La Plateforme API.", "verifiedAt": "2026-09-08" }, { "surface": "self-host", "note": "Released as open weights under a Modified MIT license.", "verifiedAt": "2026-09-08" } ], "benchmarksNote": "Mistral's API reference (docs.mistral.ai/api) does not publish a max output token figure separate from context window: it states only that prompt tokens plus max_tokens must not exceed the model's context length, so the output cap recorded here equals the published context window." }, { "slug": "mistral-small-4", "providerId": "mistral", "displayName": "Mistral Small 4", "apiModelIds": [ "mistral-small-2603" ], "releasedOn": "2026-03-16", "family": "mistral-small", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "contextWindow": 256000, "maxOutputTokens": 256000, "license": "open-weights", "lifecycle": { "status": "active", "verifiedAt": "2026-09-08", "source": "https://docs.mistral.ai/models" }, "pricing": { "inputPerMTok": 0.15, "outputPerMTok": 0.6, "verifiedAt": "2026-09-08", "source": "https://mistral.ai/pricing/api" }, "governance": { "trainsOnApiDataByDefault": true, "retentionDays": "unknown", "zeroDataRetentionAvailable": "on-request", "dataResidency": [ "eu" ], "hipaaBaa": false, "soc2": true, "fedramp": "unknown", "note": "Mistral's help center states that using API calls to improve its services is on by default, with customers retaining the right to opt out at any time from the Admin panel's Privacy menu; the Vibe and API opt-out toggles are separate. Zero data retention is available on paid plans for supported stateless API calls but is not a self-service toggle: it requires a written request with a stated reason, which Mistral reviews and may approve or deny, after which it appears in Admin privacy settings. Data is hosted in the EU by default, with an explicit opt-in to a US API endpoint available. No specific retention period in days for API prompts and responses is stated on a page this audit could confirm, so it is recorded as unknown rather than the widely repeated but unverifiable 30-day figure. Mistral states it complies with SOC 2 Type II and ISO 27001/27701; a HIPAA Business Associate Agreement and any FedRAMP authorization are not confirmed on a page this audit could read, so they are recorded conservatively rather than assumed.", "verifiedAt": "2026-09-08", "sources": [ "https://docs.mistral.ai/admin/monitor-comply/zero-data-retention", "https://help.mistral.ai/en/articles/455207-can-i-opt-out-of-my-input-or-output-data-being-used-for-training", "https://help.mistral.ai/en/articles/347629-where-do-you-store-my-data-or-my-organization-s-data", "https://help.mistral.ai/en/articles/323772-do-you-have-soc2-or-iso27001-certification" ] }, "availability": [ { "surface": "direct", "modelId": "mistral-small-2603", "note": "First-party Mistral La Plateforme API.", "verifiedAt": "2026-09-08" }, { "surface": "self-host", "note": "Released under the Apache 2.0 license; weights published for self-hosting.", "verifiedAt": "2026-09-08" } ], "benchmarksNote": "Mistral's API reference (docs.mistral.ai/api) does not publish a max output token figure separate from context window: it states only that prompt tokens plus max_tokens must not exceed the model's context length, so the output cap recorded here equals the published context window." }, { "slug": "ministral-3-14b", "providerId": "mistral", "displayName": "Ministral 3 14B", "apiModelIds": [ "ministral-3-14b" ], "releasedOn": "2025-12-02", "family": "ministral-3", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "contextWindow": 256000, "maxOutputTokens": 256000, "license": "open-weights", "lifecycle": { "status": "active", "verifiedAt": "2026-09-08", "source": "https://docs.mistral.ai/models" }, "pricing": { "inputPerMTok": 0.2, "outputPerMTok": 0.2, "verifiedAt": "2026-09-08", "source": "https://mistral.ai/pricing/api" }, "governance": { "trainsOnApiDataByDefault": true, "retentionDays": "unknown", "zeroDataRetentionAvailable": "on-request", "dataResidency": [ "eu" ], "hipaaBaa": false, "soc2": true, "fedramp": "unknown", "note": "Mistral's help center states that using API calls to improve its services is on by default, with customers retaining the right to opt out at any time from the Admin panel's Privacy menu; the Vibe and API opt-out toggles are separate. Zero data retention is available on paid plans for supported stateless API calls but is not a self-service toggle: it requires a written request with a stated reason, which Mistral reviews and may approve or deny, after which it appears in Admin privacy settings. Data is hosted in the EU by default, with an explicit opt-in to a US API endpoint available. No specific retention period in days for API prompts and responses is stated on a page this audit could confirm, so it is recorded as unknown rather than the widely repeated but unverifiable 30-day figure. Mistral states it complies with SOC 2 Type II and ISO 27001/27701; a HIPAA Business Associate Agreement and any FedRAMP authorization are not confirmed on a page this audit could read, so they are recorded conservatively rather than assumed.", "verifiedAt": "2026-09-08", "sources": [ "https://docs.mistral.ai/admin/monitor-comply/zero-data-retention", "https://help.mistral.ai/en/articles/455207-can-i-opt-out-of-my-input-or-output-data-being-used-for-training", "https://help.mistral.ai/en/articles/347629-where-do-you-store-my-data-or-my-organization-s-data", "https://help.mistral.ai/en/articles/323772-do-you-have-soc2-or-iso27001-certification" ] }, "availability": [ { "surface": "direct", "modelId": "ministral-3-14b", "note": "First-party Mistral La Plateforme API.", "verifiedAt": "2026-09-08" }, { "surface": "self-host", "note": "Released under the Apache 2.0 license; weights published for self-hosting.", "verifiedAt": "2026-09-08" } ], "benchmarksNote": "Mistral's API reference (docs.mistral.ai/api) does not publish a max output token figure separate from context window: it states only that prompt tokens plus max_tokens must not exceed the model's context length, so the output cap recorded here equals the published context window." }, { "slug": "codestral", "providerId": "mistral", "displayName": "Codestral (25.08)", "apiModelIds": [ "codestral-2508" ], "releasedOn": "2025-08-01", "family": "codestral", "modalities": { "input": [ "text" ], "output": [ "text" ] }, "contextWindow": 256000, "maxOutputTokens": 256000, "license": "source-available", "lifecycle": { "status": "active", "verifiedAt": "2026-09-08", "source": "https://docs.mistral.ai/models" }, "pricing": { "inputPerMTok": 0.3, "outputPerMTok": 0.9, "cachedInputPerMTok": 0.03, "verifiedAt": "2026-09-08", "source": "https://mistral.ai/pricing/api" }, "governance": { "trainsOnApiDataByDefault": true, "retentionDays": "unknown", "zeroDataRetentionAvailable": "on-request", "dataResidency": [ "eu" ], "hipaaBaa": false, "soc2": true, "fedramp": "unknown", "note": "Mistral's help center states that using API calls to improve its services is on by default, with customers retaining the right to opt out at any time from the Admin panel's Privacy menu; the Vibe and API opt-out toggles are separate. Zero data retention is available on paid plans for supported stateless API calls but is not a self-service toggle: it requires a written request with a stated reason, which Mistral reviews and may approve or deny, after which it appears in Admin privacy settings. Data is hosted in the EU by default, with an explicit opt-in to a US API endpoint available. No specific retention period in days for API prompts and responses is stated on a page this audit could confirm, so it is recorded as unknown rather than the widely repeated but unverifiable 30-day figure. Mistral states it complies with SOC 2 Type II and ISO 27001/27701; a HIPAA Business Associate Agreement and any FedRAMP authorization are not confirmed on a page this audit could read, so they are recorded conservatively rather than assumed.", "verifiedAt": "2026-09-08", "sources": [ "https://docs.mistral.ai/admin/monitor-comply/zero-data-retention", "https://help.mistral.ai/en/articles/455207-can-i-opt-out-of-my-input-or-output-data-being-used-for-training", "https://help.mistral.ai/en/articles/347629-where-do-you-store-my-data-or-my-organization-s-data", "https://help.mistral.ai/en/articles/323772-do-you-have-soc2-or-iso27001-certification" ] }, "availability": [ { "surface": "direct", "modelId": "codestral-2508", "note": "First-party Mistral La Plateforme API.", "verifiedAt": "2026-09-08" } ], "benchmarksNote": "Mistral's API reference (docs.mistral.ai/api) does not publish a max output token figure separate from context window: it states only that prompt tokens plus max_tokens must not exceed the model's context length, so the output cap recorded here equals the published context window. Released under the Mistral Non-Production License (MNPL-0.1): free for research, testing, and internal evaluation, with commercial production use requiring a separate Mistral commercial license, which is why this model is recorded as source-available rather than open-weights." }, { "slug": "deepseek-v4-flash", "providerId": "deepseek", "displayName": "DeepSeek V4 Flash", "apiModelIds": [ "deepseek-v4-flash", "deepseek-v4-flash-0731" ], "releasedOn": "2026-07-31", "family": "deepseek-v4", "modalities": { "input": [ "text" ], "output": [ "text" ] }, "contextWindow": 1000000, "maxOutputTokens": 384000, "license": "open-weights", "lifecycle": { "status": "active", "verifiedAt": "2026-09-08", "source": "https://api-docs.deepseek.com/updates/" }, "pricing": { "inputPerMTok": 0.44, "outputPerMTok": 1.32, "cachedInputPerMTok": 0.014, "verifiedAt": "2026-09-08", "source": "https://api-docs.deepseek.com/quick_start/pricing" }, "governance": { "trainsOnApiDataByDefault": true, "retentionDays": "unknown", "zeroDataRetentionAvailable": "no", "dataResidency": [ "cn" ], "hipaaBaa": false, "soc2": false, "fedramp": "none", "note": "DeepSeek's privacy policy states personal data, including prompts, is directly collected, processed, and stored in the People's Republic of China, and may also be stored on servers outside a user's home country. Users have a stated right to opt out of having their data used for training or optimizing DeepSeek's technologies; the policy does not describe this as off by default for API traffic, so training is recorded as on by default here unless a customer opts out. No specific retention period in days is published, and no zero data retention offering, HIPAA BAA, SOC 2, or FedRAMP authorization is stated.", "verifiedAt": "2026-09-08", "sources": [ "https://cdn.deepseek.com/policies/en-US/deepseek-privacy-policy.html" ] }, "availability": [ { "surface": "direct", "modelId": "deepseek-v4-flash", "note": "First-party DeepSeek API.", "verifiedAt": "2026-09-08" }, { "surface": "self-host", "note": "Weights published on Hugging Face under the MIT license.", "verifiedAt": "2026-09-08" } ], "benchmarksNote": "DeepSeek runs a standing off-peak discount on top of the standard rate recorded here: prices are halved on cache-hit input, cache-miss input, and output tokens outside peak hours of 01:00 to 04:00 and 06:00 to 10:00 UTC, Monday through Friday. The price fields in this dataset record the standard (peak) rate; the off-peak halving is not a separately documented pricing field." }, { "slug": "deepseek-v4-pro", "providerId": "deepseek", "displayName": "DeepSeek V4 Pro", "apiModelIds": [ "deepseek-v4-pro", "deepseek-v4-pro-0813" ], "releasedOn": "2026-08-13", "family": "deepseek-v4", "modalities": { "input": [ "text" ], "output": [ "text" ] }, "contextWindow": 1000000, "maxOutputTokens": 384000, "license": "open-weights", "lifecycle": { "status": "active", "verifiedAt": "2026-09-08", "source": "https://api-docs.deepseek.com/updates/" }, "pricing": { "inputPerMTok": 1.32, "outputPerMTok": 3.96, "cachedInputPerMTok": 0.044, "verifiedAt": "2026-09-08", "source": "https://api-docs.deepseek.com/quick_start/pricing" }, "governance": { "trainsOnApiDataByDefault": true, "retentionDays": "unknown", "zeroDataRetentionAvailable": "no", "dataResidency": [ "cn" ], "hipaaBaa": false, "soc2": false, "fedramp": "none", "note": "DeepSeek's privacy policy states personal data, including prompts, is directly collected, processed, and stored in the People's Republic of China, and may also be stored on servers outside a user's home country. Users have a stated right to opt out of having their data used for training or optimizing DeepSeek's technologies; the policy does not describe this as off by default for API traffic, so training is recorded as on by default here unless a customer opts out. No specific retention period in days is published, and no zero data retention offering, HIPAA BAA, SOC 2, or FedRAMP authorization is stated.", "verifiedAt": "2026-09-08", "sources": [ "https://cdn.deepseek.com/policies/en-US/deepseek-privacy-policy.html" ] }, "availability": [ { "surface": "direct", "modelId": "deepseek-v4-pro", "note": "First-party DeepSeek API.", "verifiedAt": "2026-09-08" }, { "surface": "self-host", "note": "Weights published on Hugging Face under the MIT license.", "verifiedAt": "2026-09-08" } ], "benchmarksNote": "DeepSeek runs a standing off-peak discount on top of the standard rate recorded here: prices are halved on cache-hit input, cache-miss input, and output tokens outside peak hours of 01:00 to 04:00 and 06:00 to 10:00 UTC, Monday through Friday. The price fields in this dataset record the standard (peak) rate; the off-peak halving is not a separately documented pricing field." }, { "slug": "deepseek-chat", "providerId": "deepseek", "displayName": "DeepSeek Chat (legacy alias)", "apiModelIds": [ "deepseek-chat" ], "family": "deepseek-v3", "modalities": { "input": [ "text" ], "output": [ "text" ] }, "license": "open-weights", "lifecycle": { "status": "retired", "sunsetOn": "2026-07-24", "replacedBy": "deepseek-v4-flash", "verifiedAt": "2026-09-08", "source": "https://api-docs.deepseek.com/updates/" }, "pricing": { "inputPerMTok": 0.44, "outputPerMTok": 1.32, "cachedInputPerMTok": 0.014, "verifiedAt": "2026-09-08", "source": "https://api-docs.deepseek.com/quick_start/pricing" }, "governance": { "trainsOnApiDataByDefault": true, "retentionDays": "unknown", "zeroDataRetentionAvailable": "no", "dataResidency": [ "cn" ], "hipaaBaa": false, "soc2": false, "fedramp": "none", "note": "DeepSeek's privacy policy states personal data, including prompts, is directly collected, processed, and stored in the People's Republic of China, and may also be stored on servers outside a user's home country. Users have a stated right to opt out of having their data used for training or optimizing DeepSeek's technologies; the policy does not describe this as off by default for API traffic, so training is recorded as on by default here unless a customer opts out. No specific retention period in days is published, and no zero data retention offering, HIPAA BAA, SOC 2, or FedRAMP authorization is stated.", "verifiedAt": "2026-09-08", "sources": [ "https://cdn.deepseek.com/policies/en-US/deepseek-privacy-policy.html" ] }, "availability": [ { "surface": "direct", "note": "Retired name; requests using deepseek-chat stopped resolving at 15:59 UTC on 2026-07-24. During the transition window it pointed to the non-thinking mode of deepseek-v4-flash.", "verifiedAt": "2026-09-08" } ], "benchmarksNote": "deepseek-chat was the alias for DeepSeek-V3's non-thinking mode before the V4 migration. Its per-model documentation page was withdrawn on retirement, so context window, max output tokens, and release date have no first-party source and are not published here." }, { "slug": "deepseek-reasoner", "providerId": "deepseek", "displayName": "DeepSeek Reasoner (legacy alias)", "apiModelIds": [ "deepseek-reasoner" ], "family": "deepseek-r1", "modalities": { "input": [ "text" ], "output": [ "text" ] }, "license": "open-weights", "lifecycle": { "status": "retired", "sunsetOn": "2026-07-24", "replacedBy": "deepseek-v4-flash", "verifiedAt": "2026-09-08", "source": "https://api-docs.deepseek.com/updates/" }, "pricing": { "inputPerMTok": 0.44, "outputPerMTok": 1.32, "cachedInputPerMTok": 0.014, "verifiedAt": "2026-09-08", "source": "https://api-docs.deepseek.com/quick_start/pricing" }, "governance": { "trainsOnApiDataByDefault": true, "retentionDays": "unknown", "zeroDataRetentionAvailable": "no", "dataResidency": [ "cn" ], "hipaaBaa": false, "soc2": false, "fedramp": "none", "note": "DeepSeek's privacy policy states personal data, including prompts, is directly collected, processed, and stored in the People's Republic of China, and may also be stored on servers outside a user's home country. Users have a stated right to opt out of having their data used for training or optimizing DeepSeek's technologies; the policy does not describe this as off by default for API traffic, so training is recorded as on by default here unless a customer opts out. No specific retention period in days is published, and no zero data retention offering, HIPAA BAA, SOC 2, or FedRAMP authorization is stated.", "verifiedAt": "2026-09-08", "sources": [ "https://cdn.deepseek.com/policies/en-US/deepseek-privacy-policy.html" ] }, "availability": [ { "surface": "direct", "note": "Retired name; requests using deepseek-reasoner stopped resolving at 15:59 UTC on 2026-07-24. During the transition window it pointed to the thinking mode of deepseek-v4-flash.", "verifiedAt": "2026-09-08" } ], "benchmarksNote": "deepseek-reasoner was the alias for DeepSeek-R1's reasoning mode before the V4 migration. Its per-model documentation page was withdrawn on retirement, so context window, max output tokens, and release date have no first-party source and are not published here." }, { "slug": "grok-4-6", "providerId": "xai", "displayName": "Grok 4.6", "apiModelIds": [ "grok-4.6" ], "releasedOn": "2026-08-12", "family": "grok-4", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "contextWindow": 500000, "maxOutputTokens": 500000, "license": "proprietary", "lifecycle": { "status": "active", "verifiedAt": "2026-09-08", "source": "https://docs.x.ai/developers/release-notes" }, "pricing": { "inputPerMTok": 2, "outputPerMTok": 6, "cachedInputPerMTok": 0.5, "verifiedAt": "2026-09-08", "source": "https://docs.x.ai/developers/pricing" }, "governance": { "trainsOnApiDataByDefault": false, "retentionDays": 30, "zeroDataRetentionAvailable": "yes", "dataResidency": [ "us" ], "hipaaBaa": true, "soc2": true, "fedramp": "unknown", "note": "xAI's API security FAQ states that xAI never trains on API inputs or outputs without explicit permission, and that requests and responses are stored encrypted at rest for 30 days for abuse-auditing purposes before automatic deletion. Zero data retention is a self-serve toggle a team admin can enable from the xAI Console, applying automatically to all of that team's API keys. xAI states it is SOC 2 Type 2 compliant and offers a Business Associate Agreement on request via a BAA questionnaire; no FedRAMP authorization is published. The Data Processing Addendum states personal information is processed in the United States; this audit could read that page's existence and title but the live fetch returned a 403 during the audit, so the residency claim here rests on the DPA's indexed description rather than a direct read.", "verifiedAt": "2026-09-08", "sources": [ "https://docs.x.ai/developers/faq/security", "https://x.ai/legal/data-processing-addendum" ] }, "availability": [ { "surface": "direct", "modelId": "grok-4.6", "note": "First-party xAI API.", "verifiedAt": "2026-09-08" } ], "benchmarksNote": "docs.x.ai states Grok 4.6 has \"no text output limit\"; maxOutputTokens is recorded here as equal to the published context window, which is what that statement means in practice. Prompts at or above 200,000 tokens bill at $4 input, $1 cached input, and $12 output per million tokens instead of the standard $2, $0.50, and $6." }, { "slug": "grok-4-5", "providerId": "xai", "displayName": "Grok 4.5", "apiModelIds": [ "grok-4.5", "grok-4.5-latest", "grok-4.5-build-latest" ], "releasedOn": "2026-07-08", "family": "grok-4", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "contextWindow": 500000, "maxOutputTokens": 500000, "license": "proprietary", "lifecycle": { "status": "active", "verifiedAt": "2026-09-08", "source": "https://docs.x.ai/developers/release-notes" }, "pricing": { "inputPerMTok": 2, "outputPerMTok": 6, "cachedInputPerMTok": 0.3, "verifiedAt": "2026-09-08", "source": "https://docs.x.ai/developers/pricing" }, "governance": { "trainsOnApiDataByDefault": false, "retentionDays": 30, "zeroDataRetentionAvailable": "yes", "dataResidency": [ "us" ], "hipaaBaa": true, "soc2": true, "fedramp": "unknown", "note": "xAI's API security FAQ states that xAI never trains on API inputs or outputs without explicit permission, and that requests and responses are stored encrypted at rest for 30 days for abuse-auditing purposes before automatic deletion. Zero data retention is a self-serve toggle a team admin can enable from the xAI Console, applying automatically to all of that team's API keys. xAI states it is SOC 2 Type 2 compliant and offers a Business Associate Agreement on request via a BAA questionnaire; no FedRAMP authorization is published. The Data Processing Addendum states personal information is processed in the United States; this audit could read that page's existence and title but the live fetch returned a 403 during the audit, so the residency claim here rests on the DPA's indexed description rather than a direct read.", "verifiedAt": "2026-09-08", "sources": [ "https://docs.x.ai/developers/faq/security", "https://x.ai/legal/data-processing-addendum" ] }, "availability": [ { "surface": "direct", "modelId": "grok-4.5", "note": "First-party xAI API.", "verifiedAt": "2026-09-08" } ], "benchmarksNote": "docs.x.ai states Grok 4.5 has \"no text output limit\"; maxOutputTokens is recorded here as equal to the published context window. Prompts at or above 200,000 tokens bill at $4 input, $0.60 cached input, and $12 output per million tokens instead of the standard $2, $0.30, and $6." }, { "slug": "grok-build-0-1", "providerId": "xai", "displayName": "Grok Build 0.1", "apiModelIds": [ "grok-build-0.1" ], "releasedOn": "2026-05-19", "family": "grok-build", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "contextWindow": 256000, "maxOutputTokens": 256000, "license": "proprietary", "lifecycle": { "status": "active", "verifiedAt": "2026-09-08", "source": "https://docs.x.ai/developers/release-notes" }, "pricing": { "inputPerMTok": 1, "outputPerMTok": 2, "cachedInputPerMTok": 0.2, "verifiedAt": "2026-09-08", "source": "https://docs.x.ai/developers/pricing" }, "governance": { "trainsOnApiDataByDefault": false, "retentionDays": 30, "zeroDataRetentionAvailable": "yes", "dataResidency": [ "us" ], "hipaaBaa": true, "soc2": true, "fedramp": "unknown", "note": "xAI's API security FAQ states that xAI never trains on API inputs or outputs without explicit permission, and that requests and responses are stored encrypted at rest for 30 days for abuse-auditing purposes before automatic deletion. Zero data retention is a self-serve toggle a team admin can enable from the xAI Console, applying automatically to all of that team's API keys. xAI states it is SOC 2 Type 2 compliant and offers a Business Associate Agreement on request via a BAA questionnaire; no FedRAMP authorization is published. The Data Processing Addendum states personal information is processed in the United States; this audit could read that page's existence and title but the live fetch returned a 403 during the audit, so the residency claim here rests on the DPA's indexed description rather than a direct read.", "verifiedAt": "2026-09-08", "sources": [ "https://docs.x.ai/developers/faq/security", "https://x.ai/legal/data-processing-addendum" ] }, "availability": [ { "surface": "direct", "modelId": "grok-build-0.1", "note": "First-party xAI API.", "verifiedAt": "2026-09-08" } ], "benchmarksNote": "xAI does not publish a max output token figure for this model separate from context window on its model-card page; maxOutputTokens is recorded here as equal to the published context window, consistent with the explicit \"no text output limit\" statement on Grok 4.5 and Grok 4.6's model cards. Currently in early access, trained specifically for agentic coding workflows. Prompts at or above 200,000 tokens bill at $2 input, $0.40 cached input, and $4 output per million tokens instead of the standard $1, $0.20, and $2. The retired grok-code-fast-1 alias now redirects here." }, { "slug": "grok-4-20-reasoning", "providerId": "xai", "displayName": "Grok 4.20 (reasoning)", "apiModelIds": [ "grok-4.20-0309-reasoning", "grok-4.20", "grok-4.20-reasoning", "grok-4.20-reasoning-latest" ], "releasedOn": "2026-03-16", "family": "grok-4.20", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "contextWindow": 1000000, "maxOutputTokens": 1000000, "license": "proprietary", "lifecycle": { "status": "active", "verifiedAt": "2026-09-08", "source": "https://docs.x.ai/developers/release-notes" }, "pricing": { "inputPerMTok": 1.25, "outputPerMTok": 2.5, "cachedInputPerMTok": 0.2, "verifiedAt": "2026-09-08", "source": "https://docs.x.ai/developers/pricing" }, "governance": { "trainsOnApiDataByDefault": false, "retentionDays": 30, "zeroDataRetentionAvailable": "yes", "dataResidency": [ "us" ], "hipaaBaa": true, "soc2": true, "fedramp": "unknown", "note": "xAI's API security FAQ states that xAI never trains on API inputs or outputs without explicit permission, and that requests and responses are stored encrypted at rest for 30 days for abuse-auditing purposes before automatic deletion. Zero data retention is a self-serve toggle a team admin can enable from the xAI Console, applying automatically to all of that team's API keys. xAI states it is SOC 2 Type 2 compliant and offers a Business Associate Agreement on request via a BAA questionnaire; no FedRAMP authorization is published. The Data Processing Addendum states personal information is processed in the United States; this audit could read that page's existence and title but the live fetch returned a 403 during the audit, so the residency claim here rests on the DPA's indexed description rather than a direct read.", "verifiedAt": "2026-09-08", "sources": [ "https://docs.x.ai/developers/faq/security", "https://x.ai/legal/data-processing-addendum" ] }, "availability": [ { "surface": "direct", "modelId": "grok-4.20-0309-reasoning", "note": "First-party xAI API.", "verifiedAt": "2026-09-08" } ], "benchmarksNote": "xAI does not publish a max output token figure for this model separate from context window on its model-card page; maxOutputTokens is recorded here as equal to the published context window, consistent with the explicit \"no text output limit\" statement on Grok 4.5 and Grok 4.6's model cards. Prompts at or above 200,000 tokens bill at $2.50 input, $0.40 cached input, and $5 output per million tokens instead of the standard $1.25, $0.20, and $2.50." }, { "slug": "grok-4-20-non-reasoning", "providerId": "xai", "displayName": "Grok 4.20 (non-reasoning)", "apiModelIds": [ "grok-4.20-0309-non-reasoning", "grok-4.20-non-reasoning", "grok-4.20-non-reasoning-latest" ], "releasedOn": "2026-03-16", "family": "grok-4.20", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "contextWindow": 1000000, "maxOutputTokens": 1000000, "license": "proprietary", "lifecycle": { "status": "active", "verifiedAt": "2026-09-08", "source": "https://docs.x.ai/developers/release-notes" }, "pricing": { "inputPerMTok": 1.25, "outputPerMTok": 2.5, "cachedInputPerMTok": 0.2, "verifiedAt": "2026-09-08", "source": "https://docs.x.ai/developers/pricing" }, "governance": { "trainsOnApiDataByDefault": false, "retentionDays": 30, "zeroDataRetentionAvailable": "yes", "dataResidency": [ "us" ], "hipaaBaa": true, "soc2": true, "fedramp": "unknown", "note": "xAI's API security FAQ states that xAI never trains on API inputs or outputs without explicit permission, and that requests and responses are stored encrypted at rest for 30 days for abuse-auditing purposes before automatic deletion. Zero data retention is a self-serve toggle a team admin can enable from the xAI Console, applying automatically to all of that team's API keys. xAI states it is SOC 2 Type 2 compliant and offers a Business Associate Agreement on request via a BAA questionnaire; no FedRAMP authorization is published. The Data Processing Addendum states personal information is processed in the United States; this audit could read that page's existence and title but the live fetch returned a 403 during the audit, so the residency claim here rests on the DPA's indexed description rather than a direct read.", "verifiedAt": "2026-09-08", "sources": [ "https://docs.x.ai/developers/faq/security", "https://x.ai/legal/data-processing-addendum" ] }, "availability": [ { "surface": "direct", "modelId": "grok-4.20-0309-non-reasoning", "note": "First-party xAI API.", "verifiedAt": "2026-09-08" } ], "benchmarksNote": "xAI does not publish a max output token figure for this model separate from context window on its model-card page; maxOutputTokens is recorded here as equal to the published context window, consistent with the explicit \"no text output limit\" statement on Grok 4.5 and Grok 4.6's model cards. Prompts at or above 200,000 tokens bill at $2.50 input, $0.40 cached input, and $5 output per million tokens instead of the standard $1.25, $0.20, and $2.50." }, { "slug": "grok-4-20-multi-agent", "providerId": "xai", "displayName": "Grok 4.20 Multi-Agent", "apiModelIds": [ "grok-4.20-multi-agent-0309", "grok-4.20-multi-agent", "grok-4.20-multi-agent-latest" ], "releasedOn": "2026-03-16", "family": "grok-4.20", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "contextWindow": 1000000, "maxOutputTokens": 1000000, "license": "proprietary", "lifecycle": { "status": "active", "verifiedAt": "2026-09-08", "source": "https://docs.x.ai/developers/release-notes" }, "pricing": { "inputPerMTok": 1.25, "outputPerMTok": 2.5, "cachedInputPerMTok": 0.2, "verifiedAt": "2026-09-08", "source": "https://docs.x.ai/developers/pricing" }, "governance": { "trainsOnApiDataByDefault": false, "retentionDays": 30, "zeroDataRetentionAvailable": "yes", "dataResidency": [ "us" ], "hipaaBaa": true, "soc2": true, "fedramp": "unknown", "note": "xAI's API security FAQ states that xAI never trains on API inputs or outputs without explicit permission, and that requests and responses are stored encrypted at rest for 30 days for abuse-auditing purposes before automatic deletion. Zero data retention is a self-serve toggle a team admin can enable from the xAI Console, applying automatically to all of that team's API keys. xAI states it is SOC 2 Type 2 compliant and offers a Business Associate Agreement on request via a BAA questionnaire; no FedRAMP authorization is published. The Data Processing Addendum states personal information is processed in the United States; this audit could read that page's existence and title but the live fetch returned a 403 during the audit, so the residency claim here rests on the DPA's indexed description rather than a direct read.", "verifiedAt": "2026-09-08", "sources": [ "https://docs.x.ai/developers/faq/security", "https://x.ai/legal/data-processing-addendum" ] }, "availability": [ { "surface": "direct", "modelId": "grok-4.20-multi-agent-0309", "note": "First-party xAI API.", "verifiedAt": "2026-09-08" } ], "benchmarksNote": "xAI does not publish a max output token figure for this model separate from context window on its model-card page; maxOutputTokens is recorded here as equal to the published context window, consistent with the explicit \"no text output limit\" statement on Grok 4.5 and Grok 4.6's model cards. Runs xAI's 16-agent Heavy system. Prompts at or above 200,000 tokens bill at $2.50 input, $0.40 cached input, and $5 output per million tokens instead of the standard $1.25, $0.20, and $2.50." }, { "slug": "grok-4-1-fast-reasoning", "providerId": "xai", "displayName": "Grok 4.1 Fast (reasoning, retired)", "apiModelIds": [ "grok-4-1-fast-reasoning" ], "family": "grok-4.1", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "license": "proprietary", "lifecycle": { "status": "retired", "sunsetOn": "2026-05-15", "replacedBy": "grok-4-20-reasoning", "verifiedAt": "2026-09-08", "source": "https://docs.x.ai/developers/migration/may-15-retirement" }, "pricing": { "inputPerMTok": 1.25, "outputPerMTok": 2.5, "verifiedAt": "2026-09-08", "source": "https://docs.x.ai/developers/migration/may-15-retirement" }, "governance": { "trainsOnApiDataByDefault": false, "retentionDays": 30, "zeroDataRetentionAvailable": "yes", "dataResidency": [ "us" ], "hipaaBaa": true, "soc2": true, "fedramp": "unknown", "note": "xAI's API security FAQ states that xAI never trains on API inputs or outputs without explicit permission, and that requests and responses are stored encrypted at rest for 30 days for abuse-auditing purposes before automatic deletion. Zero data retention is a self-serve toggle a team admin can enable from the xAI Console, applying automatically to all of that team's API keys. xAI states it is SOC 2 Type 2 compliant and offers a Business Associate Agreement on request via a BAA questionnaire; no FedRAMP authorization is published. The Data Processing Addendum states personal information is processed in the United States; this audit could read that page's existence and title but the live fetch returned a 403 during the audit, so the residency claim here rests on the DPA's indexed description rather than a direct read.", "verifiedAt": "2026-09-08", "sources": [ "https://docs.x.ai/developers/faq/security", "https://x.ai/legal/data-processing-addendum" ] }, "availability": [ { "surface": "direct", "note": "Retired May 15, 2026, 12:00 PM PT. Requests to this slug now redirect to Grok 4.3 with low reasoning effort and bill at Grok 4.3 rates.", "verifiedAt": "2026-09-08" } ], "benchmarksNote": "Its per-model documentation page was withdrawn on retirement, so context window, max output tokens, and release date have no first-party source and are not published here." }, { "slug": "grok-4-1-fast-non-reasoning", "providerId": "xai", "displayName": "Grok 4.1 Fast (non-reasoning, retired)", "apiModelIds": [ "grok-4-1-fast-non-reasoning" ], "family": "grok-4.1", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "license": "proprietary", "lifecycle": { "status": "retired", "sunsetOn": "2026-05-15", "replacedBy": "grok-4-20-non-reasoning", "verifiedAt": "2026-09-08", "source": "https://docs.x.ai/developers/migration/may-15-retirement" }, "pricing": { "inputPerMTok": 1.25, "outputPerMTok": 2.5, "verifiedAt": "2026-09-08", "source": "https://docs.x.ai/developers/migration/may-15-retirement" }, "governance": { "trainsOnApiDataByDefault": false, "retentionDays": 30, "zeroDataRetentionAvailable": "yes", "dataResidency": [ "us" ], "hipaaBaa": true, "soc2": true, "fedramp": "unknown", "note": "xAI's API security FAQ states that xAI never trains on API inputs or outputs without explicit permission, and that requests and responses are stored encrypted at rest for 30 days for abuse-auditing purposes before automatic deletion. Zero data retention is a self-serve toggle a team admin can enable from the xAI Console, applying automatically to all of that team's API keys. xAI states it is SOC 2 Type 2 compliant and offers a Business Associate Agreement on request via a BAA questionnaire; no FedRAMP authorization is published. The Data Processing Addendum states personal information is processed in the United States; this audit could read that page's existence and title but the live fetch returned a 403 during the audit, so the residency claim here rests on the DPA's indexed description rather than a direct read.", "verifiedAt": "2026-09-08", "sources": [ "https://docs.x.ai/developers/faq/security", "https://x.ai/legal/data-processing-addendum" ] }, "availability": [ { "surface": "direct", "note": "Retired May 15, 2026, 12:00 PM PT. Requests to this slug now redirect to Grok 4.3 with none reasoning effort and bill at Grok 4.3 rates.", "verifiedAt": "2026-09-08" } ], "benchmarksNote": "Its per-model documentation page was withdrawn on retirement, so context window, max output tokens, and release date have no first-party source and are not published here." }, { "slug": "grok-4-fast-reasoning", "providerId": "xai", "displayName": "Grok 4 Fast (reasoning, retired)", "apiModelIds": [ "grok-4-fast-reasoning" ], "family": "grok-4", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "license": "proprietary", "lifecycle": { "status": "retired", "sunsetOn": "2026-05-15", "replacedBy": "grok-4-20-reasoning", "verifiedAt": "2026-09-08", "source": "https://docs.x.ai/developers/migration/may-15-retirement" }, "pricing": { "inputPerMTok": 1.25, "outputPerMTok": 2.5, "verifiedAt": "2026-09-08", "source": "https://docs.x.ai/developers/migration/may-15-retirement" }, "governance": { "trainsOnApiDataByDefault": false, "retentionDays": 30, "zeroDataRetentionAvailable": "yes", "dataResidency": [ "us" ], "hipaaBaa": true, "soc2": true, "fedramp": "unknown", "note": "xAI's API security FAQ states that xAI never trains on API inputs or outputs without explicit permission, and that requests and responses are stored encrypted at rest for 30 days for abuse-auditing purposes before automatic deletion. Zero data retention is a self-serve toggle a team admin can enable from the xAI Console, applying automatically to all of that team's API keys. xAI states it is SOC 2 Type 2 compliant and offers a Business Associate Agreement on request via a BAA questionnaire; no FedRAMP authorization is published. The Data Processing Addendum states personal information is processed in the United States; this audit could read that page's existence and title but the live fetch returned a 403 during the audit, so the residency claim here rests on the DPA's indexed description rather than a direct read.", "verifiedAt": "2026-09-08", "sources": [ "https://docs.x.ai/developers/faq/security", "https://x.ai/legal/data-processing-addendum" ] }, "availability": [ { "surface": "direct", "note": "Retired May 15, 2026, 12:00 PM PT. Requests to this slug now redirect to Grok 4.3 with low reasoning effort and bill at Grok 4.3 rates.", "verifiedAt": "2026-09-08" } ], "benchmarksNote": "Its per-model documentation page was withdrawn on retirement, so context window, max output tokens, and release date have no first-party source and are not published here." }, { "slug": "grok-4-fast-non-reasoning", "providerId": "xai", "displayName": "Grok 4 Fast (non-reasoning, retired)", "apiModelIds": [ "grok-4-fast-non-reasoning" ], "family": "grok-4", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "license": "proprietary", "lifecycle": { "status": "retired", "sunsetOn": "2026-05-15", "replacedBy": "grok-4-20-non-reasoning", "verifiedAt": "2026-09-08", "source": "https://docs.x.ai/developers/migration/may-15-retirement" }, "pricing": { "inputPerMTok": 1.25, "outputPerMTok": 2.5, "verifiedAt": "2026-09-08", "source": "https://docs.x.ai/developers/migration/may-15-retirement" }, "governance": { "trainsOnApiDataByDefault": false, "retentionDays": 30, "zeroDataRetentionAvailable": "yes", "dataResidency": [ "us" ], "hipaaBaa": true, "soc2": true, "fedramp": "unknown", "note": "xAI's API security FAQ states that xAI never trains on API inputs or outputs without explicit permission, and that requests and responses are stored encrypted at rest for 30 days for abuse-auditing purposes before automatic deletion. Zero data retention is a self-serve toggle a team admin can enable from the xAI Console, applying automatically to all of that team's API keys. xAI states it is SOC 2 Type 2 compliant and offers a Business Associate Agreement on request via a BAA questionnaire; no FedRAMP authorization is published. The Data Processing Addendum states personal information is processed in the United States; this audit could read that page's existence and title but the live fetch returned a 403 during the audit, so the residency claim here rests on the DPA's indexed description rather than a direct read.", "verifiedAt": "2026-09-08", "sources": [ "https://docs.x.ai/developers/faq/security", "https://x.ai/legal/data-processing-addendum" ] }, "availability": [ { "surface": "direct", "note": "Retired May 15, 2026, 12:00 PM PT. Requests to this slug now redirect to Grok 4.3 with none reasoning effort and bill at Grok 4.3 rates.", "verifiedAt": "2026-09-08" } ], "benchmarksNote": "Its per-model documentation page was withdrawn on retirement, so context window, max output tokens, and release date have no first-party source and are not published here." }, { "slug": "grok-4-0709", "providerId": "xai", "displayName": "Grok 4 (0709, retired)", "apiModelIds": [ "grok-4-0709" ], "family": "grok-4", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "license": "proprietary", "lifecycle": { "status": "retired", "sunsetOn": "2026-05-15", "replacedBy": "grok-4-20-reasoning", "verifiedAt": "2026-09-08", "source": "https://docs.x.ai/developers/migration/may-15-retirement" }, "pricing": { "inputPerMTok": 1.25, "outputPerMTok": 2.5, "verifiedAt": "2026-09-08", "source": "https://docs.x.ai/developers/migration/may-15-retirement" }, "governance": { "trainsOnApiDataByDefault": false, "retentionDays": 30, "zeroDataRetentionAvailable": "yes", "dataResidency": [ "us" ], "hipaaBaa": true, "soc2": true, "fedramp": "unknown", "note": "xAI's API security FAQ states that xAI never trains on API inputs or outputs without explicit permission, and that requests and responses are stored encrypted at rest for 30 days for abuse-auditing purposes before automatic deletion. Zero data retention is a self-serve toggle a team admin can enable from the xAI Console, applying automatically to all of that team's API keys. xAI states it is SOC 2 Type 2 compliant and offers a Business Associate Agreement on request via a BAA questionnaire; no FedRAMP authorization is published. The Data Processing Addendum states personal information is processed in the United States; this audit could read that page's existence and title but the live fetch returned a 403 during the audit, so the residency claim here rests on the DPA's indexed description rather than a direct read.", "verifiedAt": "2026-09-08", "sources": [ "https://docs.x.ai/developers/faq/security", "https://x.ai/legal/data-processing-addendum" ] }, "availability": [ { "surface": "direct", "note": "Retired May 15, 2026, 12:00 PM PT. Requests to this slug now redirect to Grok 4.3 with low reasoning effort and bill at Grok 4.3 rates.", "verifiedAt": "2026-09-08" } ], "benchmarksNote": "Its per-model documentation page was withdrawn on retirement, so context window, max output tokens, and release date have no first-party source and are not published here." }, { "slug": "grok-code-fast-1", "providerId": "xai", "displayName": "Grok Code Fast 1 (retired)", "apiModelIds": [ "grok-code-fast-1" ], "family": "grok-code", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "license": "proprietary", "lifecycle": { "status": "retired", "sunsetOn": "2026-05-15", "replacedBy": "grok-build-0-1", "verifiedAt": "2026-09-08", "source": "https://docs.x.ai/developers/migration/may-15-retirement" }, "pricing": { "inputPerMTok": 1, "outputPerMTok": 2, "verifiedAt": "2026-09-08", "source": "https://docs.x.ai/developers/migration/may-15-retirement" }, "governance": { "trainsOnApiDataByDefault": false, "retentionDays": 30, "zeroDataRetentionAvailable": "yes", "dataResidency": [ "us" ], "hipaaBaa": true, "soc2": true, "fedramp": "unknown", "note": "xAI's API security FAQ states that xAI never trains on API inputs or outputs without explicit permission, and that requests and responses are stored encrypted at rest for 30 days for abuse-auditing purposes before automatic deletion. Zero data retention is a self-serve toggle a team admin can enable from the xAI Console, applying automatically to all of that team's API keys. xAI states it is SOC 2 Type 2 compliant and offers a Business Associate Agreement on request via a BAA questionnaire; no FedRAMP authorization is published. The Data Processing Addendum states personal information is processed in the United States; this audit could read that page's existence and title but the live fetch returned a 403 during the audit, so the residency claim here rests on the DPA's indexed description rather than a direct read.", "verifiedAt": "2026-09-08", "sources": [ "https://docs.x.ai/developers/faq/security", "https://x.ai/legal/data-processing-addendum" ] }, "availability": [ { "surface": "direct", "note": "Retired May 15, 2026, 12:00 PM PT. Requests to this slug now redirect to Grok Build 0.1, xAI's agentic coding model, and bill at its rates.", "verifiedAt": "2026-09-08" } ], "benchmarksNote": "Its per-model documentation page was withdrawn on retirement, so context window, max output tokens, and release date have no first-party source and are not published here." }, { "slug": "grok-3", "providerId": "xai", "displayName": "Grok 3 (retired)", "apiModelIds": [ "grok-3" ], "family": "grok-3", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "license": "proprietary", "lifecycle": { "status": "retired", "sunsetOn": "2026-05-15", "replacedBy": "grok-4-20-non-reasoning", "verifiedAt": "2026-09-08", "source": "https://docs.x.ai/developers/migration/may-15-retirement" }, "pricing": { "inputPerMTok": 1.25, "outputPerMTok": 2.5, "verifiedAt": "2026-09-08", "source": "https://docs.x.ai/developers/migration/may-15-retirement" }, "governance": { "trainsOnApiDataByDefault": false, "retentionDays": 30, "zeroDataRetentionAvailable": "yes", "dataResidency": [ "us" ], "hipaaBaa": true, "soc2": true, "fedramp": "unknown", "note": "xAI's API security FAQ states that xAI never trains on API inputs or outputs without explicit permission, and that requests and responses are stored encrypted at rest for 30 days for abuse-auditing purposes before automatic deletion. Zero data retention is a self-serve toggle a team admin can enable from the xAI Console, applying automatically to all of that team's API keys. xAI states it is SOC 2 Type 2 compliant and offers a Business Associate Agreement on request via a BAA questionnaire; no FedRAMP authorization is published. The Data Processing Addendum states personal information is processed in the United States; this audit could read that page's existence and title but the live fetch returned a 403 during the audit, so the residency claim here rests on the DPA's indexed description rather than a direct read.", "verifiedAt": "2026-09-08", "sources": [ "https://docs.x.ai/developers/faq/security", "https://x.ai/legal/data-processing-addendum" ] }, "availability": [ { "surface": "direct", "note": "Retired May 15, 2026, 12:00 PM PT. Requests to this slug now redirect to Grok 4.3 with none reasoning effort and bill at Grok 4.3 rates.", "verifiedAt": "2026-09-08" } ], "benchmarksNote": "Its per-model documentation page was withdrawn on retirement, so context window, max output tokens, and release date have no first-party source and are not published here." }, { "slug": "qwen3-8-max", "providerId": "qwen", "displayName": "Qwen3.8-Max", "apiModelIds": [ "qwen3.8-max", "qwen3.8-max-0902", "qwen3.8-max-2026-09-02" ], "releasedOn": "2026-09-02", "family": "qwen3.8", "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "contextWindow": 1000000, "maxOutputTokens": 1000000, "license": "proprietary", "lifecycle": { "status": "active", "verifiedAt": "2026-09-08", "source": "https://www.alibabacloud.com/help/en/model-studio/newly-released-models" }, "pricing": { "inputPerMTok": 2, "outputPerMTok": 6, "verifiedAt": "2026-09-08", "source": "https://www.alibabacloud.com/help/en/model-studio/model-pricing" }, "governance": { "trainsOnApiDataByDefault": false, "retentionDays": "unknown", "zeroDataRetentionAvailable": "no", "dataResidency": [ "singapore", "china", "japan", "eu", "us" ], "hipaaBaa": false, "soc2": true, "fedramp": "unknown", "note": "Alibaba Cloud's Model Studio privacy notice states it strictly protects data privacy and will never use customer data for model training, and that Model Studio does not use customer business data to develop or improve models without explicit consent. The notice states Model Studio has achieved a SOC 2 unqualified opinion covering Security, Availability, and Confidentiality; it does not mention HIPAA, ISO certifications, or FedRAMP, and does not state a specific retention period in days for logged prompts and completions. Model Studio operates regional endpoints in Singapore, China (Beijing and Hong Kong), Japan (Tokyo), Germany (Frankfurt), and the US (Virginia); no zero-data-retention toggle is documented for the API.", "verifiedAt": "2026-09-08", "sources": [ "https://www.alibabacloud.com/help/en/model-studio/privacy-notice", "https://www.alibabacloud.com/help/en/model-studio/what-is-model-studio" ] }, "availability": [ { "surface": "direct", "modelId": "qwen3.8-max", "note": "First-party Alibaba Cloud Model Studio API (DashScope).", "verifiedAt": "2026-09-08" } ], "benchmarksNote": "Alibaba Cloud's Model Studio does not publish a max output token figure for this model separate from context window; the DashScope API reference states the max_tokens parameter's default and maximum are both \"the model's maximum output length\" without giving that length elsewhere, so maxOutputTokens is recorded here as equal to the published context window." }, { "slug": "qwen3-7-plus", "providerId": "qwen", "displayName": "Qwen3.7-Plus", "apiModelIds": [ "qwen3.7-plus" ], "releasedOn": "2026-06-01", "family": "qwen3.7", "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "contextWindow": 1000000, "maxOutputTokens": 1000000, "license": "proprietary", "lifecycle": { "status": "active", "verifiedAt": "2026-09-08", "source": "https://www.alibabacloud.com/help/en/model-studio/newly-released-models" }, "pricing": { "inputPerMTok": 0.4, "outputPerMTok": 1.6, "verifiedAt": "2026-09-08", "source": "https://www.alibabacloud.com/help/en/model-studio/model-pricing" }, "governance": { "trainsOnApiDataByDefault": false, "retentionDays": "unknown", "zeroDataRetentionAvailable": "no", "dataResidency": [ "singapore", "china", "japan", "eu", "us" ], "hipaaBaa": false, "soc2": true, "fedramp": "unknown", "note": "Alibaba Cloud's Model Studio privacy notice states it strictly protects data privacy and will never use customer data for model training, and that Model Studio does not use customer business data to develop or improve models without explicit consent. The notice states Model Studio has achieved a SOC 2 unqualified opinion covering Security, Availability, and Confidentiality; it does not mention HIPAA, ISO certifications, or FedRAMP, and does not state a specific retention period in days for logged prompts and completions. Model Studio operates regional endpoints in Singapore, China (Beijing and Hong Kong), Japan (Tokyo), Germany (Frankfurt), and the US (Virginia); no zero-data-retention toggle is documented for the API.", "verifiedAt": "2026-09-08", "sources": [ "https://www.alibabacloud.com/help/en/model-studio/privacy-notice", "https://www.alibabacloud.com/help/en/model-studio/what-is-model-studio" ] }, "availability": [ { "surface": "direct", "modelId": "qwen3.7-plus", "note": "First-party Alibaba Cloud Model Studio API (DashScope).", "verifiedAt": "2026-09-08" } ], "benchmarksNote": "Alibaba Cloud's Model Studio does not publish a max output token figure for this model separate from context window; the DashScope API reference states the max_tokens parameter's default and maximum are both \"the model's maximum output length\" without giving that length elsewhere, so maxOutputTokens is recorded here as equal to the published context window. Prompts above 256,000 input tokens bill at $1.20 input and $4.80 output per million tokens (non-thinking mode) instead of the standard $0.40 and $1.60." }, { "slug": "qwen3-8-flash", "providerId": "qwen", "displayName": "Qwen3.8-Flash", "apiModelIds": [ "qwen3.8-flash" ], "releasedOn": "2026-08-26", "family": "qwen3.8", "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "contextWindow": 1000000, "maxOutputTokens": 1000000, "license": "proprietary", "lifecycle": { "status": "active", "verifiedAt": "2026-09-08", "source": "https://www.alibabacloud.com/help/en/model-studio/newly-released-models" }, "pricing": { "inputPerMTok": 0.15, "outputPerMTok": 0.47, "verifiedAt": "2026-09-08", "source": "https://www.alibabacloud.com/help/en/model-studio/model-pricing" }, "governance": { "trainsOnApiDataByDefault": false, "retentionDays": "unknown", "zeroDataRetentionAvailable": "no", "dataResidency": [ "singapore", "china", "japan", "eu", "us" ], "hipaaBaa": false, "soc2": true, "fedramp": "unknown", "note": "Alibaba Cloud's Model Studio privacy notice states it strictly protects data privacy and will never use customer data for model training, and that Model Studio does not use customer business data to develop or improve models without explicit consent. The notice states Model Studio has achieved a SOC 2 unqualified opinion covering Security, Availability, and Confidentiality; it does not mention HIPAA, ISO certifications, or FedRAMP, and does not state a specific retention period in days for logged prompts and completions. Model Studio operates regional endpoints in Singapore, China (Beijing and Hong Kong), Japan (Tokyo), Germany (Frankfurt), and the US (Virginia); no zero-data-retention toggle is documented for the API.", "verifiedAt": "2026-09-08", "sources": [ "https://www.alibabacloud.com/help/en/model-studio/privacy-notice", "https://www.alibabacloud.com/help/en/model-studio/what-is-model-studio" ] }, "availability": [ { "surface": "direct", "modelId": "qwen3.8-flash", "note": "First-party Alibaba Cloud Model Studio API (DashScope).", "verifiedAt": "2026-09-08" } ], "benchmarksNote": "Alibaba Cloud's Model Studio does not publish a max output token figure for this model separate from context window; the DashScope API reference states the max_tokens parameter's default and maximum are both \"the model's maximum output length\" without giving that length elsewhere, so maxOutputTokens is recorded here as equal to the published context window." }, { "slug": "qwen3-coder-plus", "providerId": "qwen", "displayName": "Qwen3-Coder-Plus", "apiModelIds": [ "qwen3-coder-plus" ], "releasedOn": "2025-07-23", "family": "qwen3-coder", "modalities": { "input": [ "text" ], "output": [ "text" ] }, "contextWindow": 1000000, "maxOutputTokens": 1000000, "license": "proprietary", "lifecycle": { "status": "active", "verifiedAt": "2026-09-08", "source": "https://www.alibabacloud.com/help/en/model-studio/newly-released-models" }, "pricing": { "inputPerMTok": 1, "outputPerMTok": 5, "verifiedAt": "2026-09-08", "source": "https://www.alibabacloud.com/help/en/model-studio/model-pricing" }, "governance": { "trainsOnApiDataByDefault": false, "retentionDays": "unknown", "zeroDataRetentionAvailable": "no", "dataResidency": [ "singapore", "china", "japan", "eu", "us" ], "hipaaBaa": false, "soc2": true, "fedramp": "unknown", "note": "Alibaba Cloud's Model Studio privacy notice states it strictly protects data privacy and will never use customer data for model training, and that Model Studio does not use customer business data to develop or improve models without explicit consent. The notice states Model Studio has achieved a SOC 2 unqualified opinion covering Security, Availability, and Confidentiality; it does not mention HIPAA, ISO certifications, or FedRAMP, and does not state a specific retention period in days for logged prompts and completions. Model Studio operates regional endpoints in Singapore, China (Beijing and Hong Kong), Japan (Tokyo), Germany (Frankfurt), and the US (Virginia); no zero-data-retention toggle is documented for the API.", "verifiedAt": "2026-09-08", "sources": [ "https://www.alibabacloud.com/help/en/model-studio/privacy-notice", "https://www.alibabacloud.com/help/en/model-studio/what-is-model-studio" ] }, "availability": [ { "surface": "direct", "modelId": "qwen3-coder-plus", "note": "First-party Alibaba Cloud Model Studio API (DashScope).", "verifiedAt": "2026-09-08" } ], "benchmarksNote": "Alibaba Cloud's Model Studio does not publish a max output token figure for this model separate from context window; the DashScope API reference states the max_tokens parameter's default and maximum are both \"the model's maximum output length\" without giving that length elsewhere, so maxOutputTokens is recorded here as equal to the published context window. Tiered pricing rises with prompt length: $1.80 input / $9 output per million tokens from 32K to 128K prompt tokens, $3 / $15 from 128K to 256K, and $6 / $60 from 256K to 1M, versus the standard $1 / $5 rate below 32K." }, { "slug": "qwen3-vl-plus", "providerId": "qwen", "displayName": "Qwen3-VL-Plus", "apiModelIds": [ "qwen3-vl-plus" ], "releasedOn": "2025-09-23", "family": "qwen3-vl", "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "contextWindow": 256000, "maxOutputTokens": 256000, "license": "proprietary", "lifecycle": { "status": "active", "verifiedAt": "2026-09-08", "source": "https://www.alibabacloud.com/help/en/model-studio/newly-released-models" }, "pricing": { "inputPerMTok": 0.2, "outputPerMTok": 1.6, "verifiedAt": "2026-09-08", "source": "https://www.alibabacloud.com/help/en/model-studio/model-pricing" }, "governance": { "trainsOnApiDataByDefault": false, "retentionDays": "unknown", "zeroDataRetentionAvailable": "no", "dataResidency": [ "singapore", "china", "japan", "eu", "us" ], "hipaaBaa": false, "soc2": true, "fedramp": "unknown", "note": "Alibaba Cloud's Model Studio privacy notice states it strictly protects data privacy and will never use customer data for model training, and that Model Studio does not use customer business data to develop or improve models without explicit consent. The notice states Model Studio has achieved a SOC 2 unqualified opinion covering Security, Availability, and Confidentiality; it does not mention HIPAA, ISO certifications, or FedRAMP, and does not state a specific retention period in days for logged prompts and completions. Model Studio operates regional endpoints in Singapore, China (Beijing and Hong Kong), Japan (Tokyo), Germany (Frankfurt), and the US (Virginia); no zero-data-retention toggle is documented for the API.", "verifiedAt": "2026-09-08", "sources": [ "https://www.alibabacloud.com/help/en/model-studio/privacy-notice", "https://www.alibabacloud.com/help/en/model-studio/what-is-model-studio" ] }, "availability": [ { "surface": "direct", "modelId": "qwen3-vl-plus", "note": "First-party Alibaba Cloud Model Studio API (DashScope).", "verifiedAt": "2026-09-08" } ], "benchmarksNote": "Alibaba Cloud's Model Studio does not publish a max output token figure for this model separate from context window; the DashScope API reference states the max_tokens parameter's default and maximum are both \"the model's maximum output length\" without giving that length elsewhere, so maxOutputTokens is recorded here as equal to the published context window. Tiered pricing rises with prompt length: $0.30 input / $2.40 output per million tokens from 32K to 128K prompt tokens, and $0.60 / $4.80 from 128K to 256K, versus the standard $0.20 / $1.60 rate below 32K." }, { "slug": "nova-pro", "providerId": "amazon", "displayName": "Nova Pro", "apiModelIds": [ "amazon.nova-pro-v1:0", "us.amazon.nova-pro-v1:0", "eu.amazon.nova-pro-v1:0" ], "releasedOn": "2024-12-05", "family": "nova", "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "contextWindow": 300000, "maxOutputTokens": 5000, "license": "proprietary", "lifecycle": { "status": "active", "verifiedAt": "2026-09-08", "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-amazon-nova-pro.html" }, "pricing": { "inputPerMTok": 0.8, "outputPerMTok": 3.2, "verifiedAt": "2026-09-08", "source": "https://aws.amazon.com/bedrock/pricing/" }, "governance": { "trainsOnApiDataByDefault": false, "retentionDays": "unknown", "zeroDataRetentionAvailable": "on-request", "dataResidency": [ "us", "eu", "apac" ], "hipaaBaa": true, "soc2": true, "fedramp": "high", "note": "AWS states that prompts customers enter into Amazon foundation models, and the outputs produced in response, are not used to train the underlying Amazon models unless a customer consents, and are not shared with any model provider. Amazon Bedrock's data retention is controlled by an account- or project-level mode rather than a fixed number of days: the \"none\" mode is zero data retention, \"default\" retains data only for abuse-detection purposes with no published day count, and \"aws_review\" (required only by specific non-Amazon models on Bedrock, not by Nova) retains data for up to 30 days for human review within the AWS boundary. Zero data retention for models that otherwise require retention is available on a per-account, per-model basis by contacting an AWS account manager. Amazon Bedrock is in scope for ISO, SOC, and CSA STAR Level 2, is HIPAA eligible, supports GDPR-compliant use, and is a FedRAMP High authorized service in the AWS GovCloud (US-West) Region specifically, not across all commercial regions.", "verifiedAt": "2026-09-08", "sources": [ "https://docs.aws.amazon.com/bedrock/latest/userguide/data-retention.html", "https://aws.amazon.com/bedrock/amazon-models/privacy/", "https://aws.amazon.com/bedrock/security-compliance/" ] }, "availability": [ { "surface": "bedrock", "modelId": "amazon.nova-pro-v1:0", "note": "Served exclusively through Amazon Bedrock; there is no separate first-party direct API.", "verifiedAt": "2026-09-08" } ], "benchmarksNote": "Knowledge cutoff October 2024. Amazon's balanced multimodal model for text, image, and video input." }, { "slug": "nova-lite", "providerId": "amazon", "displayName": "Nova Lite", "apiModelIds": [ "amazon.nova-lite-v1:0", "us.amazon.nova-lite-v1:0", "eu.amazon.nova-lite-v1:0" ], "releasedOn": "2024-12-05", "family": "nova", "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "contextWindow": 300000, "maxOutputTokens": 5000, "license": "proprietary", "lifecycle": { "status": "active", "verifiedAt": "2026-09-08", "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-amazon-nova-lite.html" }, "pricing": { "inputPerMTok": 0.06, "outputPerMTok": 0.24, "verifiedAt": "2026-09-08", "source": "https://aws.amazon.com/bedrock/pricing/" }, "governance": { "trainsOnApiDataByDefault": false, "retentionDays": "unknown", "zeroDataRetentionAvailable": "on-request", "dataResidency": [ "us", "eu", "apac" ], "hipaaBaa": true, "soc2": true, "fedramp": "high", "note": "AWS states that prompts customers enter into Amazon foundation models, and the outputs produced in response, are not used to train the underlying Amazon models unless a customer consents, and are not shared with any model provider. Amazon Bedrock's data retention is controlled by an account- or project-level mode rather than a fixed number of days: the \"none\" mode is zero data retention, \"default\" retains data only for abuse-detection purposes with no published day count, and \"aws_review\" (required only by specific non-Amazon models on Bedrock, not by Nova) retains data for up to 30 days for human review within the AWS boundary. Zero data retention for models that otherwise require retention is available on a per-account, per-model basis by contacting an AWS account manager. Amazon Bedrock is in scope for ISO, SOC, and CSA STAR Level 2, is HIPAA eligible, supports GDPR-compliant use, and is a FedRAMP High authorized service in the AWS GovCloud (US-West) Region specifically, not across all commercial regions.", "verifiedAt": "2026-09-08", "sources": [ "https://docs.aws.amazon.com/bedrock/latest/userguide/data-retention.html", "https://aws.amazon.com/bedrock/amazon-models/privacy/", "https://aws.amazon.com/bedrock/security-compliance/" ] }, "availability": [ { "surface": "bedrock", "modelId": "amazon.nova-lite-v1:0", "note": "Served exclusively through Amazon Bedrock; there is no separate first-party direct API.", "verifiedAt": "2026-09-08" } ], "benchmarksNote": "Knowledge cutoff October 2024. Amazon's low-cost multimodal model for document analysis and visual question answering." }, { "slug": "nova-micro", "providerId": "amazon", "displayName": "Nova Micro", "apiModelIds": [ "amazon.nova-micro-v1:0", "us.amazon.nova-micro-v1:0", "eu.amazon.nova-micro-v1:0" ], "releasedOn": "2024-12-05", "family": "nova", "modalities": { "input": [ "text" ], "output": [ "text" ] }, "contextWindow": 128000, "maxOutputTokens": 5000, "license": "proprietary", "lifecycle": { "status": "active", "verifiedAt": "2026-09-08", "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-amazon-nova-micro.html" }, "pricing": { "inputPerMTok": 0.035, "outputPerMTok": 0.14, "verifiedAt": "2026-09-08", "source": "https://aws.amazon.com/bedrock/pricing/" }, "governance": { "trainsOnApiDataByDefault": false, "retentionDays": "unknown", "zeroDataRetentionAvailable": "on-request", "dataResidency": [ "us", "eu", "apac" ], "hipaaBaa": true, "soc2": true, "fedramp": "high", "note": "AWS states that prompts customers enter into Amazon foundation models, and the outputs produced in response, are not used to train the underlying Amazon models unless a customer consents, and are not shared with any model provider. Amazon Bedrock's data retention is controlled by an account- or project-level mode rather than a fixed number of days: the \"none\" mode is zero data retention, \"default\" retains data only for abuse-detection purposes with no published day count, and \"aws_review\" (required only by specific non-Amazon models on Bedrock, not by Nova) retains data for up to 30 days for human review within the AWS boundary. Zero data retention for models that otherwise require retention is available on a per-account, per-model basis by contacting an AWS account manager. Amazon Bedrock is in scope for ISO, SOC, and CSA STAR Level 2, is HIPAA eligible, supports GDPR-compliant use, and is a FedRAMP High authorized service in the AWS GovCloud (US-West) Region specifically, not across all commercial regions.", "verifiedAt": "2026-09-08", "sources": [ "https://docs.aws.amazon.com/bedrock/latest/userguide/data-retention.html", "https://aws.amazon.com/bedrock/amazon-models/privacy/", "https://aws.amazon.com/bedrock/security-compliance/" ] }, "availability": [ { "surface": "bedrock", "modelId": "amazon.nova-micro-v1:0", "note": "Served exclusively through Amazon Bedrock; there is no separate first-party direct API.", "verifiedAt": "2026-09-08" } ], "benchmarksNote": "Knowledge cutoff October 2024. Amazon's fastest text-only model, optimized for summarization, translation, and classification." }, { "slug": "nova-premier", "providerId": "amazon", "displayName": "Nova Premier", "apiModelIds": [ "amazon.nova-premier-v1:0", "us.amazon.nova-premier-v1:0" ], "releasedOn": "2025-10-31", "family": "nova", "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "contextWindow": 1000000, "maxOutputTokens": 25000, "license": "proprietary", "lifecycle": { "status": "legacy", "sunsetNotBefore": "2026-10-31", "sunsetOn": "2026-09-14", "replacedBy": "nova-2-lite", "verifiedAt": "2026-09-08", "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-amazon-nova-premier.html" }, "pricing": { "inputPerMTok": 2.5, "outputPerMTok": 12.5, "verifiedAt": "2026-09-08", "source": "https://aws.amazon.com/bedrock/pricing/" }, "governance": { "trainsOnApiDataByDefault": false, "retentionDays": "unknown", "zeroDataRetentionAvailable": "on-request", "dataResidency": [ "us", "eu", "apac" ], "hipaaBaa": true, "soc2": true, "fedramp": "high", "note": "AWS states that prompts customers enter into Amazon foundation models, and the outputs produced in response, are not used to train the underlying Amazon models unless a customer consents, and are not shared with any model provider. Amazon Bedrock's data retention is controlled by an account- or project-level mode rather than a fixed number of days: the \"none\" mode is zero data retention, \"default\" retains data only for abuse-detection purposes with no published day count, and \"aws_review\" (required only by specific non-Amazon models on Bedrock, not by Nova) retains data for up to 30 days for human review within the AWS boundary. Zero data retention for models that otherwise require retention is available on a per-account, per-model basis by contacting an AWS account manager. Amazon Bedrock is in scope for ISO, SOC, and CSA STAR Level 2, is HIPAA eligible, supports GDPR-compliant use, and is a FedRAMP High authorized service in the AWS GovCloud (US-West) Region specifically, not across all commercial regions.", "verifiedAt": "2026-09-08", "sources": [ "https://docs.aws.amazon.com/bedrock/latest/userguide/data-retention.html", "https://aws.amazon.com/bedrock/amazon-models/privacy/", "https://aws.amazon.com/bedrock/security-compliance/" ] }, "availability": [ { "surface": "bedrock", "modelId": "amazon.nova-premier-v1:0", "note": "Served exclusively through Amazon Bedrock; there is no separate first-party direct API.", "verifiedAt": "2026-09-08" } ], "benchmarksNote": "Knowledge cutoff October 2024. Amazon's model for complex reasoning, agentic workflows, and model distillation. Its Bedrock model card lists both an EOL-no-sooner-than date of October 31, 2026 and a Model EOL date of September 14, 2026; the earlier, more specific EOL date is used here as sunsetOn. Entered Legacy status March 13, 2026." }, { "slug": "nova-2-lite", "providerId": "amazon", "displayName": "Nova 2 Lite", "apiModelIds": [ "amazon.nova-2-lite-v1:0", "us.amazon.nova-2-lite-v1:0", "eu.amazon.nova-2-lite-v1:0", "jp.amazon.nova-2-lite-v1:0", "global.amazon.nova-2-lite-v1:0" ], "releasedOn": "2025-12-02", "family": "nova-2", "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "contextWindow": 1000000, "maxOutputTokens": 64000, "license": "proprietary", "lifecycle": { "status": "active", "verifiedAt": "2026-09-08", "source": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-amazon-nova-2-lite.html" }, "pricing": { "inputPerMTok": 0.3, "outputPerMTok": 2.5, "verifiedAt": "2026-09-08", "source": "https://aws.amazon.com/bedrock/pricing/" }, "governance": { "trainsOnApiDataByDefault": false, "retentionDays": "unknown", "zeroDataRetentionAvailable": "on-request", "dataResidency": [ "us", "eu", "apac" ], "hipaaBaa": true, "soc2": true, "fedramp": "high", "note": "AWS states that prompts customers enter into Amazon foundation models, and the outputs produced in response, are not used to train the underlying Amazon models unless a customer consents, and are not shared with any model provider. Amazon Bedrock's data retention is controlled by an account- or project-level mode rather than a fixed number of days: the \"none\" mode is zero data retention, \"default\" retains data only for abuse-detection purposes with no published day count, and \"aws_review\" (required only by specific non-Amazon models on Bedrock, not by Nova) retains data for up to 30 days for human review within the AWS boundary. Zero data retention for models that otherwise require retention is available on a per-account, per-model basis by contacting an AWS account manager. Amazon Bedrock is in scope for ISO, SOC, and CSA STAR Level 2, is HIPAA eligible, supports GDPR-compliant use, and is a FedRAMP High authorized service in the AWS GovCloud (US-West) Region specifically, not across all commercial regions.", "verifiedAt": "2026-09-08", "sources": [ "https://docs.aws.amazon.com/bedrock/latest/userguide/data-retention.html", "https://aws.amazon.com/bedrock/amazon-models/privacy/", "https://aws.amazon.com/bedrock/security-compliance/" ] }, "availability": [ { "surface": "bedrock", "modelId": "amazon.nova-2-lite-v1:0", "note": "Served exclusively through Amazon Bedrock; there is no separate first-party direct API.", "verifiedAt": "2026-09-08" } ], "benchmarksNote": "Knowledge cutoff October 2025. Amazon's cost-efficient multimodal model for simple automation, document processing, and customer support." }, { "slug": "command-r-plus-08-2024", "providerId": "cohere", "displayName": "Command R+ (08-2024)", "apiModelIds": [ "command-r-plus-08-2024" ], "releasedOn": "2024-08-30", "family": "command-r", "modalities": { "input": [ "text" ], "output": [ "text" ] }, "contextWindow": 128000, "maxOutputTokens": 4000, "license": "proprietary", "lifecycle": { "status": "active", "verifiedAt": "2026-09-08", "source": "https://docs.cohere.com/docs/deprecations" }, "pricing": { "inputPerMTok": 2.5, "outputPerMTok": 10, "verifiedAt": "2026-09-08", "source": "https://cohere.com/pricing" }, "governance": { "trainsOnApiDataByDefault": true, "retentionDays": 30, "zeroDataRetentionAvailable": "enterprise-only", "dataResidency": [ "unknown" ], "hipaaBaa": false, "soc2": true, "fedramp": "unknown", "note": "Cohere's data usage policy states that, by default, prompts and generations may be used to train Cohere's models, filtered to strip common types of personal information first; customers can opt out from the Data Controls toggle in the Cohere Platform dashboard. On the SaaS Platform, logged prompts and generations are automatically deleted after 30 days unless a legal requirement or customer contract requires longer, or usage is flagged as a potential terms violation. Zero data retention is available to enterprise customers who make additional commitments, by request to support@cohere.com; it is not a self-service toggle. Cohere's Trust Center lists ISO 27001 and ISO 42001 certification and an annual SOC 2 Type II audit; no HIPAA or FedRAMP status is stated on this policy page.", "verifiedAt": "2026-09-08", "sources": [ "https://cohere.com/data-usage-policy" ] }, "availability": [ { "surface": "direct", "modelId": "command-r-plus-08-2024", "note": "First-party Cohere API.", "verifiedAt": "2026-09-08" } ], "benchmarksNote": "Cohere's own changelog entry (\"Command models get an August refresh\") does not print a dated byline, only the model's 08-2024 designation and undated prose; releasedOn is recorded as August 30, 2024 on the corroboration of multiple secondary reports of that changelog's publish date, not a first-party dateline this audit could read directly." }, { "slug": "command-r-03-2024", "providerId": "cohere", "displayName": "Command R (03-2024, retired)", "apiModelIds": [ "command-r-03-2024" ], "family": "command-r", "modalities": { "input": [ "text" ], "output": [ "text" ] }, "contextWindow": 128000, "maxOutputTokens": 4000, "license": "proprietary", "lifecycle": { "status": "retired", "sunsetOn": "2025-09-15", "replacedBy": "command-r-plus-08-2024", "verifiedAt": "2026-09-08", "source": "https://docs.cohere.com/docs/deprecations" }, "pricing": { "inputPerMTok": 0.5, "outputPerMTok": 1.5, "verifiedAt": "2026-09-08", "source": "https://cohere.com/pricing" }, "governance": { "trainsOnApiDataByDefault": true, "retentionDays": 30, "zeroDataRetentionAvailable": "enterprise-only", "dataResidency": [ "unknown" ], "hipaaBaa": false, "soc2": true, "fedramp": "unknown", "note": "Cohere's data usage policy states that, by default, prompts and generations may be used to train Cohere's models, filtered to strip common types of personal information first; customers can opt out from the Data Controls toggle in the Cohere Platform dashboard. On the SaaS Platform, logged prompts and generations are automatically deleted after 30 days unless a legal requirement or customer contract requires longer, or usage is flagged as a potential terms violation. Zero data retention is available to enterprise customers who make additional commitments, by request to support@cohere.com; it is not a self-service toggle. Cohere's Trust Center lists ISO 27001 and ISO 42001 certification and an annual SOC 2 Type II audit; no HIPAA or FedRAMP status is stated on this policy page.", "verifiedAt": "2026-09-08", "sources": [ "https://cohere.com/data-usage-policy" ] }, "availability": [ { "surface": "direct", "note": "Retired September 15, 2025. Cohere's deprecations page also lists command-r-08-2024 and command-a-03-2025 as alternative replacements.", "verifiedAt": "2026-09-08" } ] }, { "slug": "command-r-plus-04-2024", "providerId": "cohere", "displayName": "Command R+ (04-2024, retired)", "apiModelIds": [ "command-r-plus-04-2024" ], "family": "command-r", "modalities": { "input": [ "text" ], "output": [ "text" ] }, "contextWindow": 128000, "maxOutputTokens": 4000, "license": "proprietary", "lifecycle": { "status": "retired", "sunsetOn": "2025-09-15", "replacedBy": "command-r-plus-08-2024", "verifiedAt": "2026-09-08", "source": "https://docs.cohere.com/docs/deprecations" }, "pricing": { "inputPerMTok": 3, "outputPerMTok": 15, "verifiedAt": "2026-09-08", "source": "https://cohere.com/pricing" }, "governance": { "trainsOnApiDataByDefault": true, "retentionDays": 30, "zeroDataRetentionAvailable": "enterprise-only", "dataResidency": [ "unknown" ], "hipaaBaa": false, "soc2": true, "fedramp": "unknown", "note": "Cohere's data usage policy states that, by default, prompts and generations may be used to train Cohere's models, filtered to strip common types of personal information first; customers can opt out from the Data Controls toggle in the Cohere Platform dashboard. On the SaaS Platform, logged prompts and generations are automatically deleted after 30 days unless a legal requirement or customer contract requires longer, or usage is flagged as a potential terms violation. Zero data retention is available to enterprise customers who make additional commitments, by request to support@cohere.com; it is not a self-service toggle. Cohere's Trust Center lists ISO 27001 and ISO 42001 certification and an annual SOC 2 Type II audit; no HIPAA or FedRAMP status is stated on this policy page.", "verifiedAt": "2026-09-08", "sources": [ "https://cohere.com/data-usage-policy" ] }, "availability": [ { "surface": "direct", "note": "Retired September 15, 2025. Cohere's deprecations page also lists command-a-03-2025 as an alternative replacement.", "verifiedAt": "2026-09-08" } ] }, { "slug": "command-light", "providerId": "cohere", "displayName": "Command Light (retired)", "apiModelIds": [ "command-light" ], "family": "command", "modalities": { "input": [ "text" ], "output": [ "text" ] }, "contextWindow": 4000, "maxOutputTokens": 4000, "license": "proprietary", "lifecycle": { "status": "retired", "sunsetOn": "2025-09-15", "replacedBy": "command-r-plus-08-2024", "verifiedAt": "2026-09-08", "source": "https://docs.cohere.com/docs/deprecations" }, "pricing": { "inputPerMTok": 0.3, "outputPerMTok": 0.6, "verifiedAt": "2026-09-08", "source": "https://cohere.com/pricing" }, "governance": { "trainsOnApiDataByDefault": true, "retentionDays": 30, "zeroDataRetentionAvailable": "enterprise-only", "dataResidency": [ "unknown" ], "hipaaBaa": false, "soc2": true, "fedramp": "unknown", "note": "Cohere's data usage policy states that, by default, prompts and generations may be used to train Cohere's models, filtered to strip common types of personal information first; customers can opt out from the Data Controls toggle in the Cohere Platform dashboard. On the SaaS Platform, logged prompts and generations are automatically deleted after 30 days unless a legal requirement or customer contract requires longer, or usage is flagged as a potential terms violation. Zero data retention is available to enterprise customers who make additional commitments, by request to support@cohere.com; it is not a self-service toggle. Cohere's Trust Center lists ISO 27001 and ISO 42001 certification and an annual SOC 2 Type II audit; no HIPAA or FedRAMP status is stated on this policy page.", "verifiedAt": "2026-09-08", "sources": [ "https://cohere.com/data-usage-policy" ] }, "availability": [ { "surface": "direct", "note": "Retired September 15, 2025. Cohere's deprecations page recommends command-r-08-2024 or command-a-03-2025 as replacements.", "verifiedAt": "2026-09-08" } ] }, { "slug": "command", "providerId": "cohere", "displayName": "Command (retired)", "apiModelIds": [ "command" ], "family": "command", "modalities": { "input": [ "text" ], "output": [ "text" ] }, "contextWindow": 4000, "maxOutputTokens": 4000, "license": "proprietary", "lifecycle": { "status": "retired", "sunsetOn": "2025-09-15", "replacedBy": "command-r-plus-08-2024", "verifiedAt": "2026-09-08", "source": "https://docs.cohere.com/docs/deprecations" }, "pricing": { "inputPerMTok": 1, "outputPerMTok": 2, "verifiedAt": "2026-09-08", "source": "https://cohere.com/pricing" }, "governance": { "trainsOnApiDataByDefault": true, "retentionDays": 30, "zeroDataRetentionAvailable": "enterprise-only", "dataResidency": [ "unknown" ], "hipaaBaa": false, "soc2": true, "fedramp": "unknown", "note": "Cohere's data usage policy states that, by default, prompts and generations may be used to train Cohere's models, filtered to strip common types of personal information first; customers can opt out from the Data Controls toggle in the Cohere Platform dashboard. On the SaaS Platform, logged prompts and generations are automatically deleted after 30 days unless a legal requirement or customer contract requires longer, or usage is flagged as a potential terms violation. Zero data retention is available to enterprise customers who make additional commitments, by request to support@cohere.com; it is not a self-service toggle. Cohere's Trust Center lists ISO 27001 and ISO 42001 certification and an annual SOC 2 Type II audit; no HIPAA or FedRAMP status is stated on this policy page.", "verifiedAt": "2026-09-08", "sources": [ "https://cohere.com/data-usage-policy" ] }, "availability": [ { "surface": "direct", "note": "Retired September 15, 2025. Cohere's deprecations page recommends command-r-08-2024 or command-a-03-2025 as replacements.", "verifiedAt": "2026-09-08" } ] } ] }