Claude Sonnet 4: benchmark results
Anthropic Claude Sonnet 4 model, the balanced Sonnet-tier Claude 4 row. Provider: Anthropic. Released 2025-05-22. Access: API.
Aggregate Rating 1697 ± 3, rank #199 of 1656 rated models, from 1605 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| AMemGym | 33.6 | Normalized memory score (times 100): overall question-answer | 100 |
| AMemGym (Off-Policy) | 33.9 | Normalized memory score (times 100): overall question-answer | 100 |
| ATP-Bench - Encyclopedia | 83.69 | Final Score (0-100) assigned by the Chief Judge of a Gemini | 100 |
| BanglaSocialBench - Social Customs | 86.67 | Accuracy (%) on the four-option Bengali social-customs items | 100 |
| CaptchaBench - Coordinates | 93.5 | Task success rate (%; 200 test episodes) | 100 |
| ChronosBench - Complex Scenes (Guided Prompt) | 72.22 | Task Completion Rate (%; equal-weight mean of positive and n | 100 |
| ClosureBench (Chain-of-Thought) - L2 Chain | 90.2 | Accuracy (%; lenient scoring of one JSON answer per question | 100 |
| ComBench (Compilation Error Repair, OpenHands Agent) | 42 | Semantic correctness (%) of patches for the 200 reproducible | 100 |
| ComBench (Compilation Error Repair, OpenHands Agent) - Compile Success | 65.5 | Compile success (%) on the 200 reproducible real CI compilat | 100 |
| EComStage - Intent Recognition | 90.42 | Task Score (%) | 100 |
| EComStage - RAG-QA | 72.13 | Task Score (%) | 100 |
| EstGraph - Structure Identification (BA) | 100 | Accuracy (%; share of Barabasi-Albert test graphs whose gene | 100 |
Interactive version: theaggregate.ai/model?slug=claude-sonnet-4 · How It Works · Data refreshed daily, snapshot 2026-10-20.