Claude Opus 4.6 (High): benchmark results
Claude Opus 4.6 evaluated at the high reasoning-effort setting. Provider: Anthropic. Released 2026-02-05. Access: API.
Aggregate Rating 1779 ± 7, rank #173 of 2168 rated models, from 280 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| AgentKernelArena - PyTorch-to-HIP | 6.89 | Mean speedup (x; arithmetic mean over the 24 PyTorch-to-HIP | 100 |
| CLBench Life - Creation & Revision Histories | 20 | Task solving rate (%) on the 45 creation and revision histor | 100 |
| CLBench Life - Private Conversations | 20.7 | Task solving rate (%) on the 45 private one-on-one conversat | 100 |
| CTI-REALM | 0.64 | Normalized Reward | 100 |
| DataClawBench - Easy | 76.8 | Final-answer accuracy (%) on the 131 easy tasks, OpenClaw ag | 100 |
| DataClawBench - Hard | 44.8 | Final-answer accuracy (%) on the 75 hard tasks, OpenClaw age | 100 |
| DataClawBench - Medium | 62.2 | Final-answer accuracy (%) on the 286 medium tasks, OpenClaw | 100 |
| DeepResearchBench | 55.3 | Average Score | 100 |
| Horangi 4 - Ko-MT-Bench | 90.13 | Judge rating (1-10, x10) | 100 |
| Hy-MultiTurn - Action Suppression | 64.8 | Importance-weighted score (%) on the 35 action suppression d | 100 |
| HypoArena - Social Science | 1671.5 | Rating on the 163 social science cases: Bradley-Terry-Davids | 100 |
| MathArena - Project Euler 971-984 | 92.86 | Accuracy (%, direct Project Euler problems 971-984) | 100 |
Interactive version: theaggregate.ai/model?slug=claude-opus-4-6-high · How It Works · Data refreshed daily, snapshot 2026-10-20.