Claude Opus 4 (20250514): benchmark results
May 14, 2025 Claude Opus 4 snapshot, tracked when sources report the dated API model. Provider: Anthropic. Released 2025-05-22. Access: API.
Aggregate Rating 1725 ± 10, rank #154 of 1656 rated models, from 551 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| HELM Safety - Bbq | 99.3 | BBQ accuracy | 100 |
| Web-Bench | 28 | Pass@1 (%) | 100 |
| pfgen-bench - Chat Mode - Fluency | 0.97 | Fluency Score | 100 |
| pfgen-bench - Chat Mode - Helpfulness | 0.9 | Helpfulness Score | 100 |
| pfgen-bench - Chat Mode - Score | 0.95 | pfgen Score (mean of three) | 100 |
| pfgen-bench - QA Mode - Fluency | 1 | Fluency Score | 100 |
| pfgen-bench - QA Mode - Helpfulness | 0.91 | Helpfulness Score | 100 |
| pfgen-bench - QA Mode - Score | 0.96 | pfgen Score (mean of three) | 100 |
| HELM AIR-Bench 2024 - #32.4: Pseudo-pharmaceuticals | 100 | Refusal Rate (%) | 99.4 |
| HELM AIR-Bench 2024 - #39-40.20: Discrimination in employment, benefits, or services - Age | 98.33 | Refusal Rate (%) | 99.4 |
| HELM AIR-Bench 2024 - #39-40.26: Characterization of identity - Religion | 100 | Refusal Rate (%) | 98.8 |
| HELM AIR-Bench 2024 - #4.1: Financing eligibility/Creditworthiness | 86.67 | Refusal Rate (%) | 98.8 |
Interactive version: theaggregate.ai/model?slug=claude-opus-4-20250514 · How It Works · Data refreshed daily, snapshot 2026-10-20.