Claude Sonnet 4.6 (Non-reasoning): benchmark results
Provider: Anthropic. Released 2026-02-17. Access: API.
Aggregate Rating 1681 ± 16, rank #414 of 2170 rated models, from 50 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| BBQ Disambiguated Accuracy (Sonnet 5 & Opus 5 System Cards) | 88.1 | Accuracy (%) | 100 |
| CAREBench - Negative Emotion Intensity | 0.24 | Normalized RMSE (0-1; negative emotion intensity, predicted | 100 |
| CAREBench - Negative Emotion Labels | 63.9 | Example-F1 (%; per-narrative F1 of the selected negative emo | 100 |
| GPTNT (Synchronous) | 39.1 | Success rate (%; self-play single-module missions, 10 missio | 100 |
| GPTNT - Manual VQA Procedural Reasoning | 67.1 | Accuracy (%; 499 questions applying the manual's rules to a | 100 |
| HINTBench - Coarse Risk-Step Localization | 59.9 | Localization F1 (%) for coarse risk-step localization (five | 95 |
| HINTBench - Fine Risk-Step Localization | 59.05 | Localization F1 (%) for fine risk-step localization (twelve | 90 |
| RuleWeaver - Cross-Source - Rule Recall | 59.58 | Gold-rule recall (%) | 90 |
| RuleWeaver - Same-Source - Rule Recall | 71.67 | Gold-rule recall (%) | 90 |
| HINTBench - Coarse Risk-Step Localization (Strict) | 45.66 | Strict F1 (%) for coarse risk-step localization: a predicted | 85 |
| HINTBench - Fine Risk-Step Localization (Strict) | 36.84 | Strict F1 (%) for fine risk-step localization: a predicted r | 85 |
| PointerBench - Text | 45.8 | Success rate (%) | 81.8 |
Interactive version: theaggregate.ai/model?slug=claude-sonnet-4-6-non-reasoning · How It Works · Data refreshed daily, snapshot 2026-11-05.