Claude Sonnet 4 (Thinking): benchmark results
Claude Sonnet 4 evaluated with thinking enabled. Provider: Anthropic. Released 2025-05-22. Access: API.
Aggregate Rating 1639 ± 7, rank #561 of 2168 rated models, from 341 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| GAIA2 - Adaptability | 42.1 | Score (%) | 100 |
| GosuEvals | 74.3 | Score (%) | 100 |
| LLM2014 Code 2025-07 | 66.98 | Median Score | 100 |
| MME-Safety | 86.24 | Overall safety score (%; mean of relevance rate, 1 - harmful | 100 |
| OGCaReBench (Oracle Report) | 86.7 | Accuracy (%) with the source case report of each question gi | 100 |
| OpenCompass Reasoning - Common - English (CompassBench 2507) | 76.7 | Score (%) | 100 |
| WildAgtEval | 67.5 | Complexity-Injected API Call Accuracy (imported) | 100 |
| BaxBench - No Security Reminder - Correct | 75 | Correct solutions, pass@1 (%) | 97.3 |
| SEAL - MASK | 95.33 | Score | 97 |
| BenchTable - Reasoning | 85.6 | Weighted Score (%) | 95.9 |
| LLM2014 Code 2025-09 | 61.01 | Median Score | 94.4 |
| Nejumi 4 - BFCL - Non-Live AST | 81.85 | Accuracy (%) | 94.4 |
Interactive version: theaggregate.ai/model?slug=claude-sonnet-4-thinking · How It Works · Data refreshed daily, snapshot 2026-10-20.