Moonshot AI (China) · last updated 2026-05-27
Moonshot Kimi
Kimi is Moonshot AI's chat product, distinguished early by very long context windows (initially 2M tokens, expanded subsequently) which made it the preferred Chinese AI for document analysis and long-form research. Strong adoption in Chinese knowledge-work and academic contexts. Web search grounding is integrated; the long-context capability lets Kimi consume entire retrieved documents rather than chunked snippets.
Grounding model
Kimi models with web search grounding against a proprietary index biased toward Chinese-language and Chinese-academic sources. The long-context architecture means retrieved documents are ingested in full rather than chunked, which changes what content gets cited: complete, well-structured pages are favoured over keyword-matched paragraphs.
Citation style
Sources are listed at the end of the answer (rather than inline), with title, URL, and a brief relevance note. Citation count per answer is high (often 8 to 15 sources for substantive research queries).
What it rewards
- Long-form, comprehensive content that benefits from full-document ingestion
- Chinese-language coverage, particularly academic or research-oriented
- Clean document structure (headings, sections, references)
- Citable sentences distributed across the document, not just in the first paragraph
What it penalizes
- Thin or aggressively-chunked content
- Pages blocked from Chinese crawlers
- Content without clear semantic structure