DarkSide of the Moon Open Sources Latest Kimi K2.7 Code Programming Model, High-Speed Version to Go Live Next Monday

According to Watchful AI monitoring, Moonshot AI has released and open-sourced the latest Kimi K2.7 Code programming model. In the context of long-text programming, Kimi K2.7 Code has improved instruction compliance and performance in long-range programming tasks, while reducing the tendency to overthink, resulting in a 30% reduction in average token consumption. In the Kimi Code Bench v2 benchmark test evaluating code capabilities, the new model exhibited a 21.8% performance improvement, and in benchmark tests assessing the intelligent agent's autonomous execution capabilities such as Kimi Claw 24/7 Bench, it showed a performance improvement of around 10%.

Enterprises and developers can now access Kimi K2.7 Code via the Kimi API open platform. The price per 1M tokens for standard input and output is 6.5 yuan and 27 yuan, respectively, remaining the same as the K2.6 model, while the input price for cached hits has been adjusted to 1.3 yuan. The default model in the Kimi Code Plan has been synchronously upgraded, and using Kimi K2.7 Code requires enabling the thinking mode to achieve optimal performance. Moonshot AI has also announced that an accelerated version model, with an output speed 5-6 times faster, will be launched on June 15, with an output speed of around 180 Tokens/s in normal scenarios and a calling price twice that of the regular version.