Claude Opus 4.6: benchmark results
Uses adaptive thinking to spend more reasoning on difficult steps. Its deeper checking can slow simple tasks; Anthropic recommends lowering effort when it overthinks. Provider: Anthropic. Released 2026-02-05. Access: API.
Aggregate Rating 1806 ± 2, rank #63 of 1656 rated models, from 2841 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| AA Global-MMLU-Lite - Chinese | 93.58 | Accuracy (%) | 100 |
| AA Global-MMLU-Lite - English | 95.17 | Accuracy (%) | 100 |
| AA Global-MMLU-Lite - Korean | 93 | Accuracy (%) | 100 |
| AEC-Bench (Nomic Tools) - Sheet Index Consistency | 85.5 | Mean reward (0-100) from the task verifier on the 14 sheet-i | 100 |
| AEC-Bench - Submittal Review | 17.1 | Mean reward (0-100) from the task verifier on the 36 submitt | 100 |
| AGC-Bench - poetmt | 1.55 | Dataset z-score | 100 |
| AGC-Bench - sdat | 0.68 | Dataset z-score | 100 |
| AISI Cyber Cooling Tower 100M | 1.4 | Avg Steps (/7) | 100 |
| ARC-AGI-2 Verified (Opus 4.6 System Card) | 68.8 | Score (%) | 100 |
| ARFBench - Presence | 88.3 | Accuracy (%) on the 111 Presence questions (is there an anom | 100 |
| ATLAS | 70.55 | ATLAScore (self-reported) | 100 |
| Agent2 RL-Bench - ALFWorld | 97.76 | ALFWorld success rate (%) on 134 evaluation episodes of the | 100 |
Interactive version: theaggregate.ai/model?slug=claude-opus-4-6 · How It Works · Data refreshed daily, snapshot 2026-10-20.