Claude Opus 4.7 tops Vals benchmarks

This looks like a strong release for developers who care about messy, end-to-end tasks, but it’s still benchmark leadership inside a curated eval stack, not proof of universal dominance.

• –It leads Vals’ weighted index at 71.5%, which is more interesting than a single benchmark win because it spans finance, law, and coding
• –The biggest signal for builders is agentic utility: strong results on SWE-bench, Terminal-Bench, and Vibe Code Bench suggest better multi-step execution, not just prettier answers
• –Vision also matters here: Vals has Opus 4.7 ahead on multimodal and image-heavy tasks like MortgageTax and close to the top on other visual workloads
• –It does not sweep every category, which is a reminder that model quality is still domain-specific and that competitors remain competitive in academic, legal, and healthcare evals
• –Treat this as a practical frontier-model update, but still validate on your own workload before switching production defaults