On June 15th, Kimi unveiled the high-speed iteration of its K2.7 Code model, extending its availability to members of the Kimi Code Beta initiative, API creators, and Kimi Business clientele. While the foundational logic of the model stays intact, its output velocity has undergone a remarkable boost, now operating at a pace 5 to 6 times quicker than before. For tasks involving brief contexts, the processing speed can soar up to an impressive 260 Tokens per second, maintaining a steady rate of approximately 180 Tokens per second for routine programming endeavors. In terms of pricing, the high-speed variant commands double the cost of its standard counterpart, with input and output rates pegged at 13 yuan and 54 yuan per million Tokens, respectively. However, should caching be employed, the input cost diminishes to a mere 2.6 yuan per million Tokens.

Introduced on June 12th, the Kimi K2.7 Code model is tailored specifically for programming tasks that demand handling long contexts, showcasing enhanced prowess in adhering to instructions and tackling extensive programming challenges. It adeptly mitigates the problem of excessive deliberation when navigating through intricate code logic, leading to a notable reduction in average Token consumption by roughly 30%.