Across 26 shared benchmarks, Kimi K2 Instruct scores higher on 2 and Qwen3.6 35B A3B on 24. The widest gap is browsecomp, where Qwen3.6 35B A3B scores 48.7 against 14.1. Qwen3.6 35B A3B is the cheaper of the two on tracked API pricing ($0.38 against $0.60 per million input tokens).
Best tracked score per model per benchmark (default configuration; source-attributed). ↓ marks lower-is-better metrics. Open either model for its full surface, provenance and pricing. Quoted rates are the price-setter row we currently track for each model — its direct or vendor-official listing where one exists (direct, alibaba-official), otherwise the lowest tracked offer.