Kimi K2.7-Code cuts thinking tokens 30% — but practitioners say the benchmarks don't check out

Moonshot AI released open-source Kimi K2.7-Code, claiming leaner reasoning with 30% fewer thinking tokens and double-digit gains on its proprietary benchmarks, while keeping an OpenAI-compatible API. Independent KernelBench-Hard tests found it more honest but not more capable than K2.6, with some regressions and failures, and it has not been submitted to DeepSWE.