Claude Opus 4.6 (Medium): benchmark results
Claude Opus 4.6 evaluated at the medium reasoning-effort setting. Provider: Anthropic. Released 2026-02-05. Access: API.
Aggregate Rating 1743 ± 16, rank #263 of 2170 rated models, from 36 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| BeQu - Experiment 1 - Entailment F1 | 47.2 | Entailment F1 (%) | 100 |
| MageBench S1 | 1747 | Combined 1v1 Elo | 100 |
| MageBench Season 1 | 1747 | Rating (imported) | 100 |
| MedCase-Structured (1-Shot) | 56.82 | Diagnostic accuracy (%; the model reads a case as a terminol | 100 |
| HalluWorld - Grid Memory | 3.8 | Hallucination rate (%) on the memory probes (tracking states | 96.4 |
| BeQu - Experiment 1 - Entailment Recall | 39.2 | Entailment Recall (%) | 94.7 |
| NVIDIA ComputeEval - Math Libs | 91.1 | Pass@1 (%, zero-shot, 2026.1 release, 101 problems) | 94.7 |
| NVIDIA ComputeEval - cuDNN | 36.3 | Pass@1 (%, zero-shot, 2026.1 release, 113 problems) | 94.7 |
| NVIDIA ComputeEval | 68.6 | Pass@1 (%, zero-shot, 2026.1 release, 566 problems) | 89.5 |
| DeepResearchBench | 53.2 | Average Score | 87.5 |
| Context Arena | 73.06 | Average Score (%) | 87 |
| NVIDIA ComputeEval - cuBLAS | 96.3 | Pass@1 (%, zero-shot, 2026.1 release, 81 problems) | 86.8 |
Interactive version: theaggregate.ai/model?slug=claude-opus-4-6-medium · How It Works · Data refreshed daily, snapshot 2026-11-05.