Across 7 shared benchmarks, Claude 3.5 Sonnet scores higher on 0 and GPT-6 Astra on 7. The widest gap is AA Intelligence, where GPT-6 Astra scores 53 against 10. Claude 3.5 Sonnet is the cheaper of the two on tracked API pricing ($3.00 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, direct), otherwise the lowest tracked offer.