Claude Sonnet 4.5 (Thinking): benchmark results
Claude Sonnet 4.5 evaluated with extended thinking enabled for harder reasoning and agentic tasks. Provider: Anthropic. Released 2025-09-29. Access: API.
Aggregate Rating 1676 ± 5, rank #433 of 2168 rated models, from 954 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| AgentGym2 - Data Analysis | 39.77 | Task success (%; Avg@3 on the 57 end-to-end data analysis ta | 100 |
| CodeSpecBench-Repo | 20.2 | Pass rate (%), the share of generated executable pre/postcon | 100 |
| CodeSpecBench-Repo - Correctness | 37.4 | Correctness (%), the share of generated executable pre/postc | 100 |
| EuroEval Croatian Common Sense Reasoning | 82.86 | Common Sense Reasoning Average Score (%) | 100 |
| EuroEval Czech Common Sense Reasoning | 85.14 | Common Sense Reasoning Average Score (%) | 100 |
| EuroEval Danish NLU - Angry Tweets | 64.26 | Matthews correlation coefficient (%) | 100 |
| EuroEval Hungarian Knowledge | 85.53 | Knowledge Average Score (%) | 100 |
| EuroEval Icelandic NLU - MIM-GOLD NER | 86.91 | Micro F1 excluding MISC (%) | 100 |
| EuroEval Latvian Knowledge | 87.47 | Knowledge Average Score (%) | 100 |
| EuroEval Lithuanian Common Sense Reasoning | 85.65 | Common Sense Reasoning Average Score (%) | 100 |
| EuroEval Polish Common Sense Reasoning | 80.96 | Common Sense Reasoning Average Score (%) | 100 |
| EuroEval Romanian Common Sense Reasoning | 80.3 | Common Sense Reasoning Average Score (%) | 100 |
Interactive version: theaggregate.ai/model?slug=claude-sonnet-4-5-thinking · How It Works · Data refreshed daily, snapshot 2026-10-20.