China's Moonshot AI just shipped the biggest open model ever built: 2.8 trillion parameters, a 1 million token context window, and it took the number one spot on Frontend Code Arena. Here are the four ways to use it.
Go to kimi.com and start chatting. You get search, slides, deep research, docs and sheets built in, plus agent mode. No install, no card.
02
02For builders
Put it in your terminal
Kimi Code runs K3 in your terminal like a coding agent, and there is a Kimi Work desktop app. Best for long refactors, because the whole repo fits in context.
03
03$3 / 1M in
Call the API
Through the Moonshot API or OpenRouter. It is $3 per million input and $15 per million output tokens, with the full 1M token window and image input.
04
04July 27
Run the weights yourself
Moonshot is publishing the full weights under a Modified MIT license on July 27, 2026. That means you can download it, run it locally, and nobody can take it away.
Point any OpenAI-compatible tool at Kimi K3
$ export OPENAI_BASE_URL=https://openrouter.ai/api/v1 $ export OPENAI_MODEL=moonshotai/kimi-k3 # then run your usual agent, it just works
Want this running for you?
Build it step by step inside AI Creative OS, with me in the thread.
