Claude Opus 5: benchmark results
Anthropic's Claude Opus 5 (July 24, 2026), with a 1-million-token context and a low/medium/high effort control that trades cost against capability. Provider: Anthropic. Released 2026-07-24. Access: API.
Aggregate Rating 1934 ± 4, rank #11 of 1635 rated models, from 1196 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| AI for Education Pedagogy - Technology | 91.51 | Accuracy (%) | 100 |
| ASI-Bench (Claude Code) | 50.29 | Scientific Score (0-100) | 100 |
| ASI-Bench (Claude Code) - B1 Full Guidance | 72.29 | Scientific Score (0-100) | 100 |
| ASI-Bench (Claude Code) - B2 Method Hints | 45.8 | Scientific Score (0-100) | 100 |
| ASI-Bench (Claude Code) - B3 Goal Only | 40.7 | Scientific Score (0-100) | 100 |
| ASI-Bench (Claude Code) - B4 Goal + Distractors | 42.39 | Scientific Score (0-100) | 100 |
| AesCanvas - Contextual Suitability | 91.69 | Exact-match accuracy (%; one option label per ContextCanvas | 100 |
| AffectSim (Active Observation) | 28.37 | Macro-F1 (%; A-Obs: the view reached by the authors' two-sta | 100 |
| AffectSim (Initial View) | 25.46 | Macro-F1 (%; P-Init: the agent's initial camera pose, held f | 100 |
| AffectSim (Monitor View) | 29.27 | Macro-F1 (%; P-Mon: a fixed room-level monitoring camera; fi | 100 |
| AffectSim (Reference View) | 35.68 | Macro-F1 (%; P-Ref: an informative camera trajectory chosen | 100 |
| AgentIdeaBench - Active | 7.14 | Weighted Hypothesis Score (1-10) | 100 |
Interactive version: theaggregate.ai/model?slug=claude-opus-5 · How It Works · Data refreshed daily, snapshot 2026-10-15.