Moonshot AI just dropped a new version of its open-source multimodal coding model, Kimi K2.7 Code, and this one is fast — really fast. The HighSpeed variant cranks out about 180 tokens per second on medium-length inputs and can hit up to 260 tokens/s on short contexts, making it up to six times speedier than the standard model.
https://twitter.com/Kimi_Moonshot/status/2066467110960959833
But speed comes at a cost — literally. The API pricing for the HighSpeed version is exactly double. Input cache hits run $0.38 per million tokens, cache misses $1.90 per million, and output tokens cost a hefty $8.00 per million. For comparison, the standard Kimi K2.7 Code, which is already open-source, charges $0.19, $0.95, and $4.00 per million tokens respectively.
Don't expect to jump in right away, though. Limited compute capacity means Moonshot AI is rolling out access in batches. For now, it's open to Kimi Code test program members, Kimi API developers, and enterprise users. The company says it will expand access as it adds more compute.