Claude Opus 5 (Non-reasoning): benchmark results
Provider: Anthropic. Released 2026-07-24. Access: API.
Aggregate Rating 1760 ± 28, rank #228 of 2201 rated models, from 18 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| World Model Situational Probe Passing - Coverage | 0.72 | Feature-multiset structural agreement among probe-passing ca | 100 |
| World Model Situational Probe Passing - F1 | 0.49 | Feature-multiset structural agreement among probe-passing ca | 100 |
| Wolfram LLM Benchmarking Project | 71.3 | Correct Functionality (%) | 97.1 |
| WeirdML | 86.34 | Average Score | 92.5 |
| BunchCount - Individual Counting | 22.05 | Mean absolute error (individual instances per image, lower i | 85.7 |
| DuelLab GameBench 2 - Reasoning Variants | 69.81 | GameBench2 normalized score (0–100) | 82.1 |
| BBQ Ambiguous Accuracy (Sonnet 5 & Opus 5 System Cards) | 99.98 | Accuracy (%) | 71.4 |
| PSF-Med - Frontier Hardest 500 | 8.36 | Paraphrase Flip Rate (%) | 60 |
| PSF-Med - Frontier Hardest 500 - VinDr-CXR | 10.6 | Paraphrase Flip Rate (%) | 60 |
| PSF-Med - Frontier Hardest 500 - PadChest | 5.79 | Paraphrase Flip Rate (%) | 40 |
| BunchCount - Group Counting | 10.26 | Mean absolute error (semantic groups per image, lower is bet | 28.6 |
| BBQ Disambiguated Accuracy (Sonnet 5 & Opus 5 System Cards) | 81.6 | Accuracy (%) | 25 |
Interactive version: theaggregate.ai/model?slug=claude-opus-5-non-reasoning · How It Works · Data refreshed daily, snapshot 2026-11-12.