Across 16 shared benchmarks, GPT-6 Astra scores higher on 16 and Qwen3.6 35B A3B on 0. The widest gap is TauBench V3 - Banking, where GPT-6 Astra scores 43.1 against 9.3. Qwen3.6 35B A3B is the cheaper of the two on tracked API pricing ($0.38 against $10.00 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.