According to monitoring by Beating, Moonshot AI has released and open-sourced its latest Kimi K2.7 Code programming model. In long-context programming scenarios, Kimi K2.7 Code enhances instruction adherence and performance on long-range programming tasks while reducing overthinking tendencies, resulting in a 30% decrease in average token consumption. In the Kimi Code Bench v2 benchmark, which evaluates coding capabilities, the new model shows a performance improvement of 21.8%, and in benchmarks like Kimi Claw 24/7 Bench, which assess the autonomous execution capabilities of agents, performance has improved by approximately 10%. Enterprises and developers can now access Kimi K2.7 Code through the Kimi API open platform, with a price of 6.5 yuan for standard input and 27 yuan for output per 1M tokens, maintaining the same pricing as the K2.6 model, while the price for cached inputs has been adjusted to 1.3 yuan. The default model for Kimi Code Plan has been upgraded accordingly, and to achieve optimal performance with Kimi K2.7 Code, the thinking mode must be activated. Moonshot AI also announced that a high-speed version model, which increases output speed by 5-6 times, will be launched on June 15, with a typical output speed of around 180 Tokens/s and a calling price that is double that of the standard version.
All Comments