Codestral (25.08) vs DeepSeek V4 Flash
Facts only. This page does not pick a winner. For a picked winner, see the use case comparison on Tools.
| Codestral (25.08) | DeepSeek V4 Flash | |
|---|---|---|
| API model IDs | codestral-2508 | deepseek-v4-flash, deepseek-v4-flash-0731 |
| Context window | 256,000 | 1,000,000 |
| Lifecycle | active | active |
| License | source-available | open-weights |
| Trains on API data | Yes | Yes |
| Zero data retention | on-request | no |
| Input per 1M tokens | $0.3 (verified 2026-09-08) | $0.44 (verified 2026-09-08) |
Frequently asked questions
- Which has the larger context window, Codestral (25.08) or DeepSeek V4 Flash?
- Codestral (25.08) holds 256,000 tokens and DeepSeek V4 Flash holds 1,000,000. DeepSeek V4 Flash holds more. The context window counts the prompt, any attached files, and the reply together.
- Which is cheaper, Codestral (25.08) or DeepSeek V4 Flash?
- Per million input tokens, Codestral (25.08) was $0.3 (verified 2026-09-08) and DeepSeek V4 Flash was $0.44 (verified 2026-09-08). Codestral (25.08) is cheaper per input token. Output tokens cost $0.9 and $1.32 respectively, and output volume usually decides the bill.
- Does Codestral (25.08) or DeepSeek V4 Flash train on API data?
- Codestral (25.08) trains on API data by default, and DeepSeek V4 Flash trains on API data by default. Zero data retention is on-request for Codestral (25.08) and no for DeepSeek V4 Flash. Verified 2026-09-08 and 2026-09-08 against the providers' own documentation.
- Is Codestral (25.08) or DeepSeek V4 Flash being retired?
- Codestral (25.08) is active. DeepSeek V4 Flash is active. Lifecycle checked 2026-09-08 and 2026-09-08. Pin a dated snapshot ID rather than an alias if a retirement would break you.