Chat with Kimi K2.7 Code online

Kimi K2.7 Code is the coding specialist in Moonshot AI's Kimi line, released on 12 June 2026. Where Kimi K3 is the generalist you ask about anything, K2.7 Code is the one pointed at software: a one-trillion-parameter mixture-of-experts model with about 32 billion parameters active per token, built for agents that operate across files, tools and terminal commands over an extended session rather than answering a single question.

Two things define it. The first is scope: a 262,144-token context window, which is enough to hold a substantial slice of a real codebase — several files, the configuration, the docs and a long agent transcript — without chunking tricks or aggressive summarisation. The second is restraint. Moonshot reports that K2.7 Code cuts reasoning token usage by roughly 30% against K2.6 while scoring about 21.8% higher on the lab's own Kimi Code Bench v2, which is the unusual combination of thinking less and getting more right.

Thinking is always on, so the model reasons before each response and each tool call rather than waiting to be asked. In practice that shows up as fewer wrong turns in the middle of a long refactor: it reads the error, forms a hypothesis, and checks it, instead of guessing and rerunning. It takes images as input too, so a screenshot of a broken render or a failing dashboard can go straight into the conversation.

The weights are open under a Modified MIT licence, which matters if you care about being able to self-host later or audit what you are building on. On NinjaChat it is included in every plan: pick it from the model list and keep a whole engineering task in one thread.