月之暗面 Kimi API
月之暗面 Kimi API
Moonshot Kimi API provides ultra-long context Chinese LLM — Kimi supports 200K-1M tokens context, especially strong at long document analysis.
Earliest in China to do ultra-long context LLM (200K-1M tokens)
Ultra-long context isn't cheap
Sign up at platform.moonshot.cn for API key. OpenAI-compatible: POST /v1/chat/completions with model: "moonshot-v1-128k" + messages.
Uptime · 30-day window
About this API
Moonshot AI is an AI company founded 2023 in Beijing from Tsinghua alums, main product Kimi LLM. Differentiator is "ultra-long context" — early 2024 first to launch 1M tokens context (equivalent to a book or 100k lines of code in one prompt), enabling scenarios like long document analysis, entire codebase understanding. Kimi web app is popular among Chinese young users, hailed as "one of China's ChatGPT alternatives". API design is OpenAI-compatible (just change base_url); low switching cost. Same first-tier China LLM as Qwen / GLM / DeepSeek, differentiated in long context. But since late 2024, Qwen-Long, GPT-4o caught up to 1M context; Moonshot's differentiation narrowed, pivoting to explore AI agents and new directions.
What you can build
- 1Long document analysis (academic papers, legal contracts, full book QA)
- 2Chinese LLM apps
- 3Research assistants (multiple PDFs at once)
- 4Codebase understanding (entire repo in context)
Strengths & limitations
Strengths
- Earliest in China to do ultra-long context LLM (200K-1M tokens)
- Top-tier Chinese quality among China LLMs
- Kimi web app has large consumer user base (brand effect)
Limitations
- Ultra-long context isn't cheap
- Advanced features like function calling less mature than GPT-4
- Mainly Chinese; English scenarios weaker than GPT-4
Example request
curl https://platform.moonshot.cn/<endpoint> \
-H "Authorization: Bearer $API_KEY"
# Some providers use X-Api-Key instead — verify in the docs.Getting started
Sign up at platform.moonshot.cn for API key. OpenAI-compatible: POST /v1/chat/completions with model: "moonshot-v1-128k" + messages.
FAQ
Is 1M context actually usable?+
Yes but expensive and slow. Regular scenarios suffice with 128k; 1M suits special scenarios fitting entire books / codebases.
Technical details
- Auth type
- api_key
- Pricing
- paid
- Rate limit
- 默认 RPM/TPM 按 tier;可申请提高
- Protocols
- REST
- SDKs
- python, javascript