Opus 4.7 Vulnerability Validation Benchmarks: What We Found

HackerOne evaluated Anthropic's Opus 4.7 as the model powering an agent that validates whether reported issues are actually exploitable. Across an internal validation dataset and a CVE-based benchmark, Opus 4.7 showed modest overall accuracy gains over Opus 4.6 but a substantial precision improvement (fewer false positives and far fewer output extraction failures), especially on critical CVEs; the report highlights improved instruction-following, coherence in long workflows, multi-step reasoning, and token efficiency, and concludes that higher-precision validation can reduce analyst workload and improve remediation efficiency.