Anthropic · model
Claude Sonnet 4.5 Thinking benchmark scores
As of 2026-11-22, Claude Sonnet 4.5 Thinking (Anthropic) has measured scores on 9 benchmarks and estimated scores on 45 more.
| Benchmark | Score | Source |
|---|---|---|
| AA-GPQA Diamond | 85.5% | estimated ± 2.7 pp, high confidence |
| AA-HLE | 29.6% | estimated ± 6.0 pp, medium confidence |
| AA-SciCode | 48.9% | estimated ± 3.8 pp, high confidence |
| Artificial Analysis Intelligence Index | 19.6% | estimated ± 5.8 pp, medium confidence |
| CritPt | 8.7% | estimated ± 5.1 pp, medium confidence |
| GPQA Diamond | 85.1% | estimated ± 1.5 pp, medium confidence |
| HLE | 43.8% | estimated ± 6.1 pp, medium confidence |
| LiveCodeBench v6 | 83.2% | estimated ± 2.2 pp, low confidence |
| AA-Omniscience Accuracy | 40.9% | estimated ± 6.5 pp, medium confidence |
| SciCode | 47.2% | estimated ± 4.5 pp, medium confidence |
| AA Coding Index | 52.5% | estimated ± 5.6 pp, medium confidence |
| SWE-bench Verified | 76.7% | estimated ± 4.4 pp, high confidence |
| SWE-bench Pro | 54.8% | estimated ± 2.9 pp, medium confidence |
| MMLU-Pro | 86.2% | estimated ± 4.8 pp, high confidence |
| FailureSensorIQ | 64.2% | estimated ± 1.7 pp, high confidence |
| GPQA | 85.0% | estimated ± 1.6 pp, high confidence |
| MMLU | 94.3% | estimated ± 1.2 pp, medium confidence |
| Vals GPQA Diamond | 81.6% | measured |
| Vals LiveCodeBench | 73.0% | measured |
| Vals MMLU-Pro | 87.4% | measured |
| Chess Puzzles | 19.5% | estimated ± 12.3 pp, low confidence |
| AA MMLU-Pro | 87.9% | estimated ± 1.1 pp, high confidence |
| ARC-AGI-1 | 63.7% | measured |
| ARC-AGI-2 | 13.6% | measured |
| CursorBench 3.1 | 43.9% | estimated ± 5.8 pp, low confidence |
| CursorBench 3.2 | 52.7% | estimated ± 5.4 pp, low confidence |
| FrontierCode 1.1 Main | 23.8% | estimated ± 5.2 pp, low confidence |
| FrontierSWE v2 | 6.4% | estimated ± 9.5 pp, low confidence |
| Mystery Game Puzzles | 19.1% | estimated ± 4.7 pp, medium confidence |
| NYT Connections | 37.3% | measured |
| Vals SWE-bench | 70.0% | measured |
| VulcanBench v3 | 78.6% | estimated ± 5.6 pp, low confidence |
| CursorBench 4.0 | 30.8% | estimated ± 3.4 pp, medium confidence |
| DeepSWE | 56.3% | estimated ± 7.2 pp, low confidence |
| HLE w/o tools | 39.7% | estimated ± 4.9 pp, high confidence |
| PostTrainBench v1.1 | 15.4% | estimated ± 3.7 pp, medium confidence |
| StealthMark | 41.3% | estimated ± 10.7 pp, low confidence |
| ARC-AGI-3 | 0.4% | estimated ± 8.6 pp, low confidence |
| SkillsBench official leaderboard (BenchFlow) | 43.8% | estimated ± 6.1 pp, medium confidence |
| BinaryAudit | 35.1% | estimated ± 12.1 pp, low confidence |
| CompileBench | 84.4% | measured |
| OTelBench | 22.3% | estimated ± 6.6 pp, medium confidence |
| Vibe Code Bench | 22.6% | measured |
| FACTS Multimodal | 22.7% | estimated ± 4.0 pp, medium confidence |
| NL2Repo | 40.1% | estimated ± 4.7 pp, high confidence |
| PostTrainBench v1 | 19.0% | estimated ± 3.1 pp, medium confidence |
| AA LiveCodeBench | 86.6% | estimated ± 2.6 pp, high confidence |
| React Native Evals | 70.7% | estimated ± 3.0 pp, low confidence |
| SWE-Rebench | 42.8% | estimated ± 6.5 pp, low confidence |
| PostTrainBench | 12.9% | estimated ± 3.7 pp, medium confidence |
| LABBench2 | 81.7% | estimated ± 1.8 pp, low confidence |
| ScarfBench | 41.2% | estimated ± 12.3 pp, low confidence |
| OpenHarmony Bench v1.0 | 51.1% | estimated ± 3.2 pp, high confidence |
| BioBasicsBench | 34.3% | estimated ± 5.1 pp, low confidence |