en.Wedoany.com Reported - Moonshot AI released and open-sourced the Kimi K2.7 Code programming model on June 12. This model shows improved performance across multiple benchmarks compared to its predecessor, K2.6. In long-context programming scenarios, its instruction-following ability and long-range programming task performance have been enhanced, while average token consumption has been reduced by 30%, mitigating the tendency for overthinking.

In internal and external benchmarks evaluating code capabilities, Kimi K2.7 Code scores have significantly increased compared to K2.6: Kimi Code Bench v2 improved by 21.8%, Program-Bench by 11%, and MLS Bench Lite by 31.5%. In benchmarks measuring Agent autonomous execution capabilities, such as Kimi Claw 24/7 Bench, MCP Atlas, and MCP Mark Verified, performance improvements are approximately 10%.

Starting from the release date, enterprises and developers can access the K2.7 Code model via the Kimi API open platform. Its standard input and output prices for 1M tokens remain consistent with the K2.6 model, at 6.5 yuan and 27 yuan respectively; the input price for cache hits has been adjusted to 1.3 yuan.

The default model for Kimi Code Plan has been upgraded to K2.7 Code. Users of the Kimi membership plan and the enterprise Kimi membership plan (which includes Kimi Code Plan benefits) can also experience it. Moonshot AI advises that for non-programming tasks, the more feature-rich K2.6 model is still recommended. To achieve optimal performance with the K2.7 Code model, the thinking mode must be enabled; this mode is enabled by default in both Kimi API and Kimi Code. If the thinking mode is manually disabled, the API will return an error, and Kimi Code will fall back to the K2.6 model.

Moonshot AI also previewed the Kimi K2.7 Code High-Speed Version, which offers an output speed approximately 5 to 6 times that of the standard version. In regular programming scenarios, its output speed is about 180 tokens per second, reaching up to 260 tokens per second in short-context scenarios. The high-speed version will be available on June 15. Enterprises and developers can access it via the Kimi API open platform, with pricing set at twice that of the standard version.

Open-source link: https://huggingface.co/moonshotai/Kimi-K2.7-Code

This article is compiled by Wedoany. All AI citations must indicate the source as "Wedoany". If there is any infringement or other issues, please notify us promptly, and we will modify or delete it accordingly. Email: news@wedoany.com