Moonshot AI open-sources Kimi K2.7 coding model with big benchmark gains
Moonshot AI has open-sourced Kimi K2.7 Code, a coding-focused agentic model that improves on K2.6 by up to 31% on benchmarks while using 30% fewer reasoning tokens.
Score breakdown
Kimi K2.7 Code delivers substantial benchmark improvements over its predecessor while cutting reasoning token usage by 30%, making a capable open-weights coding model more efficient and freely accessible.
- 01Moonshot AI open-sourced Kimi K2.7 Code, described as their strongest coding model yet.
- 02The model is available via the Kimi CLI and Kimi Code.
- 03Scores 21% better than K2.6 on Kimi Code Bench.
Moonshot AI has released Kimi K2.7 Code as an open-weights, coding-focused agentic model. According to Cole Medin's video, the release was largely unexpected, with Medin describing it as having "snuck up on everyone." The model is positioned as Moonshot's strongest coding model yet and is immediately accessible through the Kimi CLI and Kimi Code tools.
The benchmark improvements over K2.6 are notable across three evaluations: 21% on Kimi Code Bench, 11% on Program Bench, and 31% on MLS Bench Light.
The benchmark improvements over K2.6 are notable across three evaluations: 21% on Kimi Code Bench, 11% on Program Bench, and 31% on MLS Bench Light. Alongside those gains, the model achieves its results with 30% fewer reasoning tokens, suggesting more efficient inference. A six-time speed mode is also described as forthcoming.
Key facts
- 01Moonshot AI open-sourced Kimi K2.7 Code, described as their strongest coding model yet.
- 02The model is available via the Kimi CLI and Kimi Code.
- 03Scores 21% better than K2.6 on Kimi Code Bench.
- 04Scores 11% better than K2.6 on Program Bench.
- 05Scores 31% better than K2.6 on MLS Bench Light.
- 06Achieves these gains using 30% fewer reasoning tokens than K2.6.
- 07A six-time speed mode is described as coming soon.
Topics
Summary and scoring are generated automatically from the original article. We always link back to the publisher and never republish images or paywalled content. Last processed Jun 19, 2026 · 10:25 UTC. How this works →