Across 33 shared benchmarks, Kimi K2 Instruct scores higher on 29 and Qwen2.5 Instruct 72B on 2, with 2 level. The widest gap is aider_polyglot, where Kimi K2 Instruct scores 60 against 7.6. Qwen2.5 Instruct 72B is the cheaper of the two on tracked API pricing ($0.47 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, direct), otherwise the lowest tracked offer.