moonshot
Moonshot AI develops the Kimi family of large language models, known for its extra-long context (262K), strong document parsing, and Chinese-language understanding. The Kimi K2.x series spans multiple tiers—K2.5 / K2.6—and was among the first models in China to champion long context, excelling at analyzing very long documents, codebases, and web content.
262K token context window; parsing of very long documents (PDF / Word / web pages); visual understanding; Function Calling; JSON mode; context caching (cache_read about 17% of the input price); OpenAI-compatible interface; Kimi agent ecosystem.
Q&A and summarization of very long documents; codebase analysis; web content extraction; contract/paper review; Chinese writing and translation; agent workflows; education and research support.
262K context (lower than the 1M offered by some competitors); no image generation; rate limits are based on account quotas; some enterprise features require contacting sales; cache hits are not guaranteed.
| Model | Input / 1M | Output / 1M | Cache Read / 1M | Per 1M |
|---|---|---|---|---|
| 1M | ||||
| 1M | ||||
| 1M | ||||
| 1M |