Claude Sonnet 4.6 (Thinking): benchmark results
Claude Sonnet 4.6 evaluated with thinking enabled. Provider: Anthropic. Released 2026-02-17. Access: API.
Aggregate Rating 1718 ± 10, rank #308 of 2168 rated models, from 194 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| ClawMark | 75.8 | Weighted score (0-100): mean over tasks of the weight-normal | 100 |
| EntLORE - Explicit Lookup (Agentic Retrieval) | 54.7 | Answer accuracy (%; L1, 469 questions answered by a stated f | 100 |
| EntLORE - Latent Organizational Reasoning (Gold Documents) | 76.3 | Answer accuracy (%; L3, 234 questions whose target relation | 100 |
| MCJudgeBench - Macro-F1 | 63.7 | Macro-F1 (%, times 100) over the yes, partial and no classes | 100 |
| Nejumi 4 - MT-Bench (Japanese) - Extraction | 99.5 | Judge rating (1-10, x10) | 100 |
| SecRespond - Baseline Risk Detection | 84.2 | Detection CAP-score (%) on the Baseline Risk (BAS) capabilit | 100 |
| SecRespond - Intrusion Entity Planning | 62.7 | Planning CAP-score (%) on the Intrusion Entity (ENT) capabil | 100 |
| Nejumi 4 - MT-Bench (Japanese) - Writing | 99.5 | Judge rating (1-10, x10) | 98.6 |
| EPIC-Bench - Feasible Path | 68.46 | Navigation path score (0-100) on feasible path planning (a p | 97.7 |
| Nejumi 4 - jaster (0-shot) - ALT English-to-Japanese | 92.56 | COMET wmt22 (x100) | 97.2 |
| Pencil Puzzle Bench - Nurikabe | 13.33 | Direct-ask Success Rate (%) | 96 |
| Nejumi 4 - HalluLens | 100 | Hallucination resistance (%) | 95.8 |
Interactive version: theaggregate.ai/model?slug=claude-sonnet-4-6-thinking · How It Works · Data refreshed daily, snapshot 2026-10-20.