Claude 3.5 Sonnet: benchmark results
Anthropic Claude 3.5 Sonnet model row for later Sonnet 3.5 results. Provider: Anthropic. Released 2024-06-20. Access: API.
Aggregate Rating 1631 ± 4, rank #350 of 1656 rated models, from 1069 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| ATAD | 59.96 | Accuracy (%) on all seven tasks (mean of the task accuracies | 100 |
| ATAD - Sentence Context Anomaly | 65.75 | Accuracy (%) on the T1 sentence context anomaly task (pick t | 100 |
| AgentLeak | 55.2 | Total Leak (imported) | 100 |
| Arabic IFEval | 75.9 | Arabic Accuracy (%) | 100 |
| BeliefShift (RAG) - Evidence Sensitivity | 63 | Evidence Sensitivity Index times 100 (-100 to 100): the rate | 100 |
| BeliefShift (Zero-Shot) - Evidence Sensitivity | 51 | Evidence Sensitivity Index times 100 (-100 to 100): the rate | 100 |
| CyberCertBench - ISA/IEC 62443 | 81.1 | Accuracy (%) on 53 ISA/IEC 62443 questions (the formal indus | 100 |
| Deception Effectiveness (Lechmazur) | 1.1 | Deception Score | 100 |
| EarthSpatialBench (CoT) - Intersect Counting | 1.06 | Mean absolute error of the object count on EarthSpatialBench | 100 |
| EmbodiedBench Habitat | 68 | Avg Score (%) | 100 |
| Factorio Learning Environment | 293206 | Production Score | 100 |
| Factorio Learning Environment - Automation Milestones | 13 | Automation milestones reached | 100 |
Interactive version: theaggregate.ai/model?slug=claude-3-5-sonnet · How It Works · Data refreshed daily, snapshot 2026-10-20.