Anthropic · model
Claude Sonnet 4.6 benchmark scores
As of 2026-10-07, Claude Sonnet 4.6 (Anthropic) has measured scores on 37 benchmarks and estimated scores on 82 more.
| Benchmark | Score | Source |
|---|---|---|
| AA-GPQA Diamond | 79.9% | measured |
| AA-HLE | 13.3% | measured |
| AA-SciCode | 52.2% | estimated ± 1.7 pp, medium confidence |
| AIME26 | 95.5% | estimated ± 0.6 pp, medium confidence |
| Artificial Analysis Intelligence Index | 24.7% | measured |
| CritPt | 0.9% | measured |
| GDPval-AA | 34.8% | estimated ± 5.1 pp, medium confidence |
| GPQA Diamond | 89.9% | estimated ± 0.1 pp, high confidence |
| HLE | 49.0% | measured |
| IFBench | 56.0% | estimated ± 6.9 pp, low confidence |
| AA-LCR | 68.3% | measured |
| LiveCodeBench v6 | 86.2% | estimated ± 8.1 pp, low confidence |
| AA-Omniscience Accuracy | 38.6% | measured |
| SciCode | 40.9% | estimated ± 2.9 pp, low confidence |
| Terminal-Bench 2.1 | 74.6% | estimated ± 6.9 pp, medium confidence |
| BrowseComp | 82.8% | estimated ± 2.2 pp, high confidence |
| HLE w/ tools | 28.1% | estimated ± 5.0 pp, low confidence |
| IFEval | 89.0% | estimated ± 1.3 pp, high confidence |
| LongBench v2 | 61.5% | estimated ± 5.9 pp, medium confidence |
| VITA-Bench | 44.9% | estimated ± 10.7 pp, low confidence |
| τ²-bench results | 79.5% | measured |
| AA Coding Index | 49.5% | estimated ± 3.3 pp, medium confidence |
| AA-MMMU-Pro | 70.6% | measured |
| APEX-Agents-AA | 32.4% | estimated ± 2.9 pp, medium confidence |
| BioMysteryBench (human-difficult) | 40.2% | estimated ± 5.9 pp, low confidence |
| DeepSearchQA | 73.4% | estimated ± 7.5 pp, low confidence |
| SWE-bench Verified | 79.6% | measured |
| AutomationBench | 28.4% | estimated ± 5.8 pp, low confidence |
| BFCL v4 | 77.9% | estimated ± 8.6 pp, low confidence |
| CyberGym | 65.2% | measured |
| JobBench | 36.9% | measured |
| SWE-bench Pro | 57.1% | estimated ± 2.1 pp, high confidence |
| τ³-bench results | 68.9% | estimated ± 8.9 pp, medium confidence |
| WideResearch | 78.4% | estimated ± 1.2 pp, medium confidence |
| AA Agentic Index | 22.4% | estimated ± 5.1 pp, medium confidence |
| MMLU-Pro | 79.2% | measured |
| FrontierMath v2 (Tier 4) | 8.3% | measured |
| FrontierMath v2 (Tiers 1-3) | 32.4% | measured |
| GPQA | 89.9% | measured |
| AA-IFBench | 41.2% | measured |
| Gert Labs | 62.9% | measured |
| AA-AnalystAgent | 26.7% | estimated ± 5.5 pp, medium confidence |
| AA EnterpriseOps-Gym | 22.0% | estimated ± 3.6 pp, low confidence |
| AA Harvey LAB | 88.9% | estimated ± 1.3 pp, medium confidence |
| ApprenticeBench | 2.0% | measured |
| ARC-AGI-1 | 86.0% | measured |
| ARC-AGI-2 | 58.3% | measured |
| CursorBench 3.1 | 48.8% | measured |
| CursorBench 3.2 | 39.1% | estimated ± 1.6 pp, low confidence |
| Terminal-Bench 3.0 | 4.4% | estimated ± 6.9 pp, low confidence |
| FrontierCode 1.1 Main | 24.3% | measured |
| FrontierSWE v2 | 9.0% | estimated ± 9.5 pp, low confidence |
| OfficeQA Pro | 50.3% | estimated ± 7.2 pp, low confidence |
| OSWorld-Verified | 72.1% | measured |
| Vals GPQA Diamond | 85.6% | measured |
| Vals LiveCodeBench | 82.1% | measured |
| Vals MMLU-Pro | 87.3% | measured |
| Vals SWE-bench | 77.4% | measured |
| Terminal-Bench 2.1 (Vals AI) | 57.3% | measured |
| VulcanBench v3 | 81.9% | estimated ± 5.6 pp, low confidence |
| AA AutomationBench | 49.9% | estimated ± 5.0 pp, medium confidence |
| GDP.pdf | 21.5% | estimated ± 4.4 pp, low confidence |
| AA ITBench | 58.9% | estimated ± 3.6 pp, low confidence |
| AA Tau3 Banking | 24.1% | estimated ± 14.6 pp, low confidence |
| AA Terminal-Bench 2.1 | 73.0% | estimated ± 4.4 pp, high confidence |
| AA Terminal-Bench 4.0 | 10.6% | estimated ± 14.0 pp, low confidence |
| CursorBench 4.0 | 29.5% | estimated ± 4.4 pp, low confidence |
| DeepSWE | 13.7% | estimated ± 3.4 pp, low confidence |
| HLE w/o tools | 33.9% | estimated ± 4.9 pp, high confidence |
| OSWorld 2.0 | 8.3% | measured |
| PostTrainBench v1.1 | 24.4% | estimated ± 2.7 pp, high confidence |
| Toolathlon-Verified | 62.8% | estimated ± 1.0 pp, low confidence |
| Vibe Code Bench | 51.5% | measured |
| CharXiv | 77.4% | measured |
| CharXiv w/o tools | 77.3% | estimated ± 2.3 pp, medium confidence |
| ExploitGym | 2.0% | estimated ± 6.9 pp, low confidence |
| C-Eval | 100.0% | estimated ± 0.7 pp, low confidence |
| Claw-Eval | 67.8% | measured |
| HMMT Nov 2025 | 93.2% | estimated ± 2.6 pp, medium confidence |
| MathVision | 84.6% | estimated ± 2.6 pp, high confidence |
| MCP Atlas | 62.2% | estimated ± 8.2 pp, medium confidence |
| MCP-Tasks | 88.3% | estimated ± 6.4 pp, low confidence |
| MMAnswerBench | 83.8% | estimated ± 1.3 pp, medium confidence |
| MMLU-Redux | 89.8% | estimated ± 0.7 pp, high confidence |
| MMMU-Pro | 77.1% | estimated ± 3.2 pp, high confidence |
| NL2Repo | 44.3% | estimated ± 4.7 pp, high confidence |
| ScreenSpot Pro | 67.0% | estimated ± 7.1 pp, medium confidence |
| SuperGPQA | 95.0% | measured |
| Terminal-Bench 2.0 | 59.1% | measured |
| Toolathlon | 51.9% | estimated ± 1.7 pp, low confidence |
| VideoMMMU | 84.6% | estimated ± 1.5 pp, high confidence |
| ERQA | 64.5% | estimated ± 3.7 pp, high confidence |
| HealthBench Hard | 1.7% | estimated ± 5.1 pp, low confidence |
| MedXpertQA (MM) | 74.0% | estimated ± 6.4 pp, low confidence |
| MMLU-Pro (Arcee) | 80.5% | estimated ± 4.0 pp, medium confidence |
| React Native Evals | 80.6% | measured |
| ResearchClawBench | 18.3% | estimated ± 1.3 pp, high confidence |
| SWE-Rebench | 60.7% | measured |
| ARC-AGI-3 | 0.0% | estimated ± 2.1 pp, medium confidence |
| FrontierMath (legacy) | 33.1% | estimated ± 0.7 pp, low confidence |
| BioMysteryBench (human-solvable) | 81.7% | estimated ± 0.6 pp, low confidence |
| FrontierCode 1.1 Extended | 26.1% | estimated ± 2.4 pp, low confidence |
| HealthBench (raw) | 19.5% | estimated ± 6.4 pp, low confidence |
| HealthBench (length-adjusted) | 53.5% | estimated ± 2.5 pp, low confidence |
| HealthBench Professional | 58.8% | estimated ± 1.3 pp, medium confidence |
| HealthBench Professional (raw) | 32.4% | estimated ± 5.0 pp, low confidence |
| HLE-Verified | 35.3% | estimated ± 0.7 pp, low confidence |
| LABBench2 | 63.7% | estimated ± 1.3 pp, low confidence |
| MMMU | 79.6% | estimated ± 0.9 pp, high confidence |
| LiveCodeBench | 86.2% | estimated ± 10.2 pp, low confidence |
| Agents' Last Exam | 24.0% | estimated ± 2.1 pp, medium confidence |
| OpenHarmony Bench | 53.5% | estimated ± 3.2 pp, high confidence |
| SimpleVQA | 65.1% | estimated ± 7.8 pp, low confidence |
| MMLU | 93.1% | estimated ± 1.1 pp, low confidence |
| MMMU-Pro w/ Python | 75.8% | estimated ± 1.6 pp, high confidence |
| MathVision w/ Python | 76.8% | estimated ± 0.9 pp, low confidence |
| RealWorldQA | 84.6% | estimated ± 1.1 pp, low confidence |
| Video-MME (with subtitle) | 86.9% | estimated ± 0.9 pp, low confidence |
| OmniDocBench 1.5 | 0.3% | estimated ± 6.4 pp, low confidence |