StepFun · model
Step 3.7 Flash benchmark scores
As of 2026-11-22, Step 3.7 Flash (StepFun) has measured scores on 24 benchmarks and estimated scores on 140 more.
| Benchmark | Score | Source |
|---|---|---|
| AA-GPQA Diamond | 80.9% | measured |
| AA-HLE | 21.4% | measured |
| AA-SciCode | 43.9% | measured |
| Artificial Analysis Intelligence Index | 19.5% | measured |
| CritPt | 2.3% | measured |
| GDPval-AA | 25.8% | measured |
| GPQA Diamond | 82.3% | estimated ± 2.4 pp, high confidence |
| HLE | 24.3% | estimated ± 8.5 pp, medium confidence |
| IFBench | 65.5% | estimated ± 6.6 pp, medium confidence |
| AA-LCR | 73.7% | measured |
| LiveCodeBench v6 | 88.1% | estimated ± 2.7 pp, high confidence |
| AA-Omniscience Accuracy | 25.8% | measured |
| SciCode | 43.9% | estimated ± 3.1 pp, high confidence |
| Terminal-Bench 2.1 | 59.5% | measured |
| BrowseComp | 75.8% | measured |
| HLE w/ tools | 47.2% | measured |
| IFEval | 92.7% | estimated ± 2.8 pp, high confidence |
| LongBench v2 | 60.8% | estimated ± 5.5 pp, medium confidence |
| VITA-Bench | 43.3% | estimated ± 6.7 pp, low confidence |
| τ²-bench results | 98.5% | measured |
| AA Coding Index | 39.6% | measured |
| AA-MMMU-Pro | 75.3% | measured |
| APEX-Agents | 19.3% | estimated ± 4.9 pp, low confidence |
| APEX-Agents-AA | 14.8% | measured |
| BioMysteryBench (human-difficult) | 27.5% | estimated ± 6.8 pp, low confidence |
| DeepSearchQA | 92.8% | measured |
| SWE-bench Verified | 77.8% | estimated ± 3.6 pp, high confidence |
| BFCL v4 | 73.6% | estimated ± 2.7 pp, high confidence |
| CyberGym | 53.7% | estimated ± 6.0 pp, medium confidence |
| JobBench | 36.0% | estimated ± 7.3 pp, low confidence |
| SkillsBench | 47.0% | estimated ± 4.1 pp, medium confidence |
| SWE-bench Pro | 56.3% | measured |
| WideResearch | 62.1% | estimated ± 3.3 pp, medium confidence |
| AA Agentic Index | 19.9% | estimated ± 3.1 pp, high confidence |
| MMLU-Pro | 85.1% | estimated ± 5.0 pp, medium confidence |
| AA-IFBench | 67.3% | measured |
| FailureSensorIQ | 64.3% | estimated ± 1.7 pp, high confidence |
| GPQA | 82.2% | estimated ± 6.1 pp, medium confidence |
| MMLU | 90.5% | estimated ± 0.9 pp, high confidence |
| Vals GPQA Diamond | 80.0% | estimated ± 3.2 pp, high confidence |
| Vals LiveCodeBench | 82.7% | estimated ± 4.7 pp, medium confidence |
| Vals MMLU-Pro | 83.1% | estimated ± 2.9 pp, high confidence |
| MMMLU | 87.3% | estimated ± 0.7 pp, medium confidence |
| MMMU | 83.6% | estimated ± 2.7 pp, high confidence |
| Chess Puzzles | 19.6% | estimated ± 12.3 pp, low confidence |
| AA MMLU-Pro | 85.0% | estimated ± 0.8 pp, high confidence |
| Gert Labs | 20.3% | measured |
| AA EnterpriseOps-Gym | 45.8% | estimated ± 2.3 pp, medium confidence |
| AA Harvey LAB v1.0 | 86.6% | estimated ± 1.7 pp, medium confidence |
| AA Tau3 Banking | 39.6% | estimated ± 2.0 pp, medium confidence |
| AA Terminal-Bench 2.1 | 81.7% | estimated ± 2.6 pp, medium confidence |
| ApprenticeBench | 9.3% | estimated ± 5.8 pp, low confidence |
| ARC-AGI-1 | 76.5% | estimated ± 7.8 pp, low confidence |
| ARC-AGI-2 | 35.0% | estimated ± 14.4 pp, low confidence |
| CursorBench 3.1 | 29.0% | estimated ± 3.4 pp, low confidence |
| CursorBench 3.2 | 52.6% | estimated ± 3.6 pp, high confidence |
| Terminal-Bench 3.0 | 5.2% | estimated ± 9.1 pp, low confidence |
| FrontierCode 1.1 Main | 13.8% | estimated ± 4.6 pp, medium confidence |
| FrontierSWE v2 | 9.2% | estimated ± 8.3 pp, medium confidence |
| MazeBench | 0.0% | estimated ± 2.7 pp, medium confidence |
| Mystery Game Puzzles | 13.8% | estimated ± 14.6 pp, low confidence |
| NYT Connections | 39.7% | measured |
| OfficeQA Pro | 50.3% | estimated ± 6.8 pp, low confidence |
| OSWorld-Verified | 72.0% | estimated ± 3.8 pp, medium confidence |
| ParseBench-1 | 59.5% | estimated ± 6.2 pp, medium confidence |
| Terminal-Bench 4.1 | 0.0% | estimated ± 10.9 pp, low confidence |
| Vals SWE-bench | 62.5% | estimated ± 5.5 pp, medium confidence |
| Terminal-Bench 2.1 (Vals AI) | 52.1% | estimated ± 7.3 pp, medium confidence |
| CWE-Bench-AA | 46.9% | estimated ± 5.2 pp, medium confidence |
| AA Cyber Index | 3.5% | estimated ± 12.4 pp, low confidence |
| DeepsecBench-AA | 5.9% | estimated ± 8.2 pp, low confidence |
| HealthBench Professional (length-adjusted, Artificial Analysis) | 51.7% | estimated ± 4.3 pp, low confidence |
| HealthBench Professional (raw, Artificial Analysis) | 54.9% | estimated ± 1.1 pp, low confidence |
| AA ITBench | 43.9% | estimated ± 4.1 pp, high confidence |
| Bug Hunt Bench | 17.2% | estimated ± 10.5 pp, low confidence |
| Chartography (tools) | 80.4% | estimated ± 4.1 pp, low confidence |
| CursorBench 4.0 | 32.8% | estimated ± 3.3 pp, medium confidence |
| CWE-bench v1 | 33.1% | estimated ± 5.7 pp, low confidence |
| DeepSWE | 43.0% | estimated ± 7.5 pp, low confidence |
| DRACO | 73.1% | estimated ± 2.0 pp, medium confidence |
| FrontierCode 2.0 | 20.1% | estimated ± 6.7 pp, low confidence |
| HLE w/o tools | 31.4% | estimated ± 4.9 pp, high confidence |
| OSWorld 2.1 | 1.3% | estimated ± 3.6 pp, medium confidence |
| OSWorld 2.1 Offline | 2.9% | estimated ± 1.9 pp, low confidence |
| PostTrainBench v1.1 | 23.5% | estimated ± 4.4 pp, medium confidence |
| SpireClimb | 5.8% | estimated ± 7.2 pp, low confidence |
| StealthMark | 41.3% | estimated ± 10.7 pp, low confidence |
| Toolathlon-Verified | 54.9% | estimated ± 3.0 pp, high confidence |
| VulcanBench Frontier v4 | 74.0% | estimated ± 6.8 pp, low confidence |
| AA AutomationBench | 41.9% | estimated ± 7.9 pp, low confidence |
| GDP.pdf | 9.7% | estimated ± 4.2 pp, medium confidence |
| AA Terminal-Bench 4.0 | 4.3% | estimated ± 11.0 pp, low confidence |
| ARC-AGI-3 | 1.3% | estimated ± 9.8 pp, medium confidence |
| ARC-AGI-3 Provider Adapter | 2.7% | estimated ± 7.3 pp, low confidence |
| cursorBench41 | 44.8% | estimated ± 3.7 pp, medium confidence |
| CWE-bench | 25.8% | estimated ± 6.2 pp, low confidence |
| MKA | 32.8% | estimated ± 4.6 pp, medium confidence |
| SkillsBench official leaderboard (BenchFlow) | 44.8% | estimated ± 6.1 pp, medium confidence |
| BinaryAudit | 35.8% | estimated ± 12.1 pp, low confidence |
| OTelBench | 22.4% | estimated ± 6.6 pp, medium confidence |
| Vibe Code Bench | 14.4% | estimated ± 10.0 pp, low confidence |
| FACTS Multimodal | 23.9% | estimated ± 4.0 pp, medium confidence |
| Hermes Bench | 66.8% | estimated ± 1.8 pp, medium confidence |
| Hermes Index | 14.5% | estimated ± 5.7 pp, low confidence |
| ParseBench-2 | 65.0% | estimated ± 6.2 pp, medium confidence |
| CharXiv | 80.0% | estimated ± 3.6 pp, high confidence |
| CharXiv w/o tools | 80.4% | estimated ± 2.5 pp, high confidence |
| ExploitGym | 0.0% | estimated ± 8.3 pp, low confidence |
| AI2D_TEST | 91.9% | estimated ± 2.0 pp, medium confidence |
| C-Eval | 91.4% | estimated ± 1.0 pp, medium confidence |
| Claw-Eval | 67.1% | measured |
| CountBench | 97.4% | estimated ± 1.6 pp, medium confidence |
| DynaMath | 85.3% | estimated ± 3.0 pp, medium confidence |
| ERQA | 64.2% | estimated ± 3.1 pp, high confidence |
| MathVision | 85.6% | estimated ± 3.1 pp, high confidence |
| MCP Atlas | 42.0% | estimated ± 9.1 pp, medium confidence |
| MCP-Tasks | 76.3% | estimated ± 7.0 pp, low confidence |
| MLVU (M-Avg) | 88.2% | estimated ± 5.0 pp, high confidence |
| MMLU-Redux | 94.2% | estimated ± 0.8 pp, medium confidence |
| MMMU-Pro | 76.5% | estimated ± 1.5 pp, high confidence |
| NL2Repo | 27.3% | estimated ± 4.7 pp, medium confidence |
| OmniDocBench 1.5 | 91.1% | estimated ± 2.0 pp, medium confidence |
| PostTrainBench v1 | 21.4% | estimated ± 4.6 pp, high confidence |
| QwenClawBench | 55.9% | estimated ± 6.0 pp, low confidence |
| RealWorldQA | 83.2% | estimated ± 2.9 pp, medium confidence |
| ScreenSpot Pro | 65.9% | estimated ± 7.5 pp, low confidence |
| SimpleVQA | 79.2% | measured |
| SuperGPQA | 66.0% | estimated ± 9.3 pp, medium confidence |
| Toolathlon | 49.5% | measured |
| V* | 95.3% | measured |
| Video-MME (w/o subtitle) | 83.3% | estimated ± 1.7 pp, medium confidence |
| Video-MME (with subtitle) | 85.3% | estimated ± 3.7 pp, medium confidence |
| VideoMMMU | 87.2% | estimated ± 10.3 pp, low confidence |
| We-Math | 87.4% | estimated ± 6.7 pp, low confidence |
| AA LiveCodeBench | 87.6% | estimated ± 4.7 pp, low confidence |
| HealthBench Hard | 16.6% | estimated ± 8.6 pp, low confidence |
| LiveCodeBench Pro | 80.8% | estimated ± 6.0 pp, low confidence |
| MedXpertQA (MM) | 72.6% | estimated ± 6.5 pp, low confidence |
| MMLU-Pro (Arcee) | 85.8% | estimated ± 4.0 pp, medium confidence |
| React Native Evals | 74.0% | estimated ± 3.2 pp, high confidence |
| ResearchClawBench | 18.2% | estimated ± 1.3 pp, high confidence |
| SWE-Rebench | 0.1% | estimated ± 8.9 pp, low confidence |
| PostTrainBench | 21.7% | estimated ± 3.7 pp, medium confidence |
| Finance Agent v2 | 55.1% | estimated ± 2.4 pp, low confidence |
| BioMysteryBench (human-solvable) | 81.7% | estimated ± 0.6 pp, low confidence |
| FrontierCode 1.1 Extended | 54.7% | estimated ± 3.0 pp, low confidence |
| HealthBench (raw) | 29.4% | estimated ± 6.0 pp, low confidence |
| HealthBench (length-adjusted) | 46.5% | estimated ± 3.4 pp, low confidence |
| HealthBench Professional | 56.7% | estimated ± 4.2 pp, low confidence |
| HealthBench Professional (raw) | 41.3% | estimated ± 4.5 pp, medium confidence |
| HLE-Verified | 20.6% | estimated ± 7.1 pp, low confidence |
| ScarfBench | 24.4% | estimated ± 9.1 pp, low confidence |
| STEMBench | 2.4% | estimated ± 0.2 pp, low confidence |
| SimpleQA | 26.1% | estimated ± 4.3 pp, medium confidence |
| Agents' Last Exam | 29.6% | estimated ± 9.7 pp, low confidence |
| BabyVision w/ Python | 87.8% | estimated ± 4.0 pp, low confidence |
| OpenHarmony Bench v1.0 | 51.0% | estimated ± 2.4 pp, medium confidence |
| MRCRv2 | 82.9% | estimated ± 11.3 pp, low confidence |
| MMLongBench-Doc | 61.1% | estimated ± 2.6 pp, low confidence |
| BioBasicsBench | 33.9% | estimated ± 2.8 pp, low confidence |
| LifeVal | 66.4% | estimated ± 1.6 pp, medium confidence |
| MMMU-Pro w/ Python | 78.5% | estimated ± 2.0 pp, high confidence |
| ExploitBench Internal Port | 0.1% | estimated ± 13.8 pp, low confidence |
| PinchBench | 90.0% | estimated ± 3.6 pp, medium confidence |