DarkMoon releases Kimi K2.7 Code High-Speed Edition, accelerating reasoning by 6x and doubling API pricing

According to DynaInsight Beating monitoring, Moonshot AI has announced the launch of the latest open-source multimodal programming mega-model Kimi K2.7 Code in a high-speed version called Kimi K2.7 Code HighSpeed. The high-speed version achieves an inference speed of around 180 tokens/s on medium-length input tasks, reaching up to 260 tokens/s in short-context scenarios, and an overall speedup of up to 6 times.

The billing rate for the HighSpeed API is twice that of the standard version. Specifically, the price for input cache hits in the high-speed version is $0.38 per million tokens, cache misses are priced at $1.90 per million tokens, and the output price is $8.00 per million tokens. In comparison, the previously released Kimi K2.7 Code standard version is billed at $0.19, $0.95, and $4.00 per million tokens for input cache hits, cache misses, and output, respectively.

Due to limited computing capacity, the high-speed version is currently being released in batches. Moonshot AI has provided restricted access to the Kimi Code testing program members, Kimi API developers, and Kimi enterprise users for the high-speed version, with plans to gradually expand access as computing capacity increases.