Kimi K2.7-Code: Faster, But Lacks Independent Benchmarks

Moonshot AI has released Kimi K2.7-Code, which now includes a HighSpeed Mode that delivers up to six times faster token throughput. What's interesting is that this Beijing-based lab has not submitted the model to any independent coding benchmarks. This means developers must rely on the company's own reported numbers when making purchasing decisions. The model is available on Hugging Face and through Moonshot's Kimi API, with pricing at $0.95 per million input tokens and $4.00 per million output tokens. The Kimi Code agent starts at $19 per month. Moonshot claims this model can reduce reasoning token usage by about 30% compared to its predecessor, K2.6. However, this efficiency claim is based on five proprietary benchmarks designed and run by Moonshot itself. The HighSpeed Mode was announced on the same day as its publication, promising speeds of around 180 tokens per second for median-length coding inputs and up to 260 tokens per second for shorter tasks. This speed gain is significant for teams using automated workflows. The bottom line is that while the new Kimi K2.7-Code offers impressive speed, the lack of independent verification means developers need to carefully consider these vendor-reported figures.