Across 22 shared benchmarks, Kimi K2 Instruct scores higher on 7 and Qwen3 VL 235B A22B Reasoning on 14, with 1 level. The widest gap is Terminal-Bench Hard, where Kimi K2 Instruct scores 23.5 against 11.4. Tracked API pricing per million tokens: Kimi K2 Instruct $0.60 in / $2.50 out, Qwen3 VL 235B A22B Reasoning $0.40 in / $4.00 out.
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.