Kimi K2.7-code Claims 30% Token Efficiency Improvement

WHO:

Techmeme - Moonshot AI is the organization releasing the Kimi K2.7-Code.

VentureBeat - Elliot Arledge is a researcher who tested K2.7-Code against K2.6.

VentureBeat - Sugumaran Balasubramaniyan is a developer who challenged Moonshot AI on benchmark choices.

WHAT:

Techmeme - Kimi K2.7-Code is available under a modified MIT license.

VentureBeat - K2.7-Code has a context window of 256K.

MarkTechPost - K2.7-Code reports a 30% lower reasoning-token usage compared to K2.6.

VentureBeat - K2.7-Code has not been submitted to DeepSWE, an independent coding benchmark.

VentureBeat - K2.6 scored 24% on DeepSWE, tied with GPT-5.4-mini.

Techmeme - Kimi K2.7-Code claims a 30% lower reasoning token usage compared to K2.6.

VentureBeat - Kimi K2.7-Code is an open-source update to the K2 coding model family.

VentureBeat - K2.7-Code is built on a trillion-parameter mixture-of-experts architecture.

VentureBeat - K2.7-Code is said to produce 21.8% gains on Kimi Code Bench v2, 11% on Program Bench, and 31.5% on MLS Bench Lite.

VentureBeat - K2.7-Code runs exclusively in thinking mode and has a fixed temperature setting of 1.0.

WHERE:

VentureBeat - K2.7-Code is deployable via vLLM or SGLang.

WHEN:

VentureBeat - K2.6 was launched in April.

WHY:

VentureBeat - Moonshot AI claims K2.7-Code addresses "overthinking," which reduces thinking-token usage.

HOW:

VentureBeat - K2.7-Code authors implementations directly rather than wrapping existing libraries.

VentureBeat - K2.7-Code can be integrated via an OpenAI-compatible API for teams already using K2.6.

(UPDATED June 13, 2026 06:21 UTC)

Kimi K2.7-Code has just been released under a modified MIT license, featuring an impressive context window of 256K. This new version also boasts a 30% reduction in reasoning-token usage compared to its predecessor, K2.6, which previously scored 24% on the independent coding benchmark DeepSWE, tying with GPT-5.4-mini. However, it's important to note that K2.7-Code has not yet been submitted for evaluation to DeepSWE.

--

(UPDATED June 13, 2026 04:21 UTC)

Moonshot AI has announced the release of Kimi K2.7-Code, an open-source update to its K2 coding model family, which boasts a 30% reduction in reasoning token usage compared to its predecessor, K2.6. Researcher Elliot Arledge has tested K2.7-Code, which is built on a trillion-parameter mixture-of-experts architecture and shows impressive performance gains across various benchmarks, including 21.8% on Kimi Code Bench v2 and 31.5% on MLS Bench Lite. The new model runs exclusively in thinking mode with a fixed temperature setting of 1.0 and can be integrated via an OpenAI-compatible API, making it accessible for teams already utilizing K2.6.