Kimi K2.7 Code: Moonshot's model built for code

The Moonshot lineup has two recent builds: Kimi K2.6 (kimi-k2.6), a general-purpose model, and Kimi K2.7 Code (kimi-k2.7-code), built specifically for development. If you write code every day, the second one is worth a try.

How a code model differs from a general-purpose one

Compared to the general-purpose model, K2.7 Code:

• Keeps project context better. It understands relationships between files and does not invent imports or methods that do not exist in your codebase.
• Is more careful with syntax. Less "almost right" code that crashes on the first run because of a small detail.
• Knows developer tool formats. Diff-style edits, stack traces, linter output and test reports are familiar to it.

Example: analyzing a failed test

curl https://api.yenisei.ru/v1/chat/completions \
  -H "Authorization: Bearer $YENISEI_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "kimi-k2.7-code",
    "messages": [
      {"role": "user", "content": "The test fails with this stack trace. Find the cause and propose a patch as a diff."}
    ]
  }'

Where the general-purpose K2.6 is still the better fit

If the task involves something besides code, such as writing API documentation in plain language, explaining the architecture to a non-technical manager, or discussing product decisions, take K2.6. The code model is good at code, but the general-purpose one is better at mixed tasks and live explanation.

How to fit it into your workflow

A convenient setup: K2.7 Code in the IDE and in CI for writing code, reviewing changes and analyzing failures. K2.6 for everything else, including communication with the team. Requests are routed by task type.

Russian in comments

Comments and commit messages in Russian come out readable, without calques. If your project documents things in Russian, the model handles it calmly, with no special prompt gymnastics required.