- Terminal Bench 2.1 86.6
- SWE-bench Pro 67.7
- DeepSWE 1.1 56.6
- NL2Repo-Bench 55.9
- FrontierSWE 73.5
- MLS-Bench-Lite 41
- PaperBench 93
- AndroidBench 75.1
- QwenSWEBench 80.7
- QwenQoderBench 58.4
- QwenReactBench 1724
- QwenSVGBench 1713
- CoWorkBench 74.8
- WorkSpaceBench 67.7
- JobBench 53.4
- SkillsBench 70.2
- Agents' Last Exam (Pass / Score) 27.0 / 52.4 27
- Automation-Bench (Pass@1) 27.3
- Toolathlon Verified (Pass@1) 72.5
- WideSearch 81.9
- Best of HLE w/ tools / HLE 56.2
- GPQA Diamond 92.6
- IFBench 82.8
- $OneMillion-Bench (expert score)
- HealthBench 60.2
- PLawBench 73.2
- PRBench-Legal 57.6
- PRBench-Finance 58.3
- MRCR v2 256K (8-needle) 92.9
- LongBench v2 66.3
- MMMU-Pro 82.3
- MathVision (without CI / with CI) 95.2 / 97.7 97.7
- BabyVision (without CI / with CI) 82.0 / 91.3 91.3
- HLE-VL (w/ Tools) 52.2
- Best of ZeroBench (Pass@5) (without CI / with CI) 24.0 / 49.0 / ZeroBench-Sub 49
- LogicVista 91.9
- HiPhO 90
- PhyX 83.5
- SLAKE 90.8
- MedXpertQA-MM 80.4
- PMC-VQA 66.2
- OSWorld-Verified 86.1
- OSWorld 2.0 (binary / partial) 19.4 / 46.7 19.4
- ScreenSpot Pro 84.5
- WebArena-Verified 66.8
- AndroidWorld 85.3
- MobileWorld 77.8
- ClawEval-MM (Pass@3 / avg) 77.2 / 74.8 74.8
- Vision2Web 69
- QwenBlenderBench 69.9
- Parametric CAD Bench
- RecreationBench 51.7
- PresentBench 79.6
- CharXiv (RQ) (without CI / with CI) 88.4 / 93.5 93.5
- OmniDocBench 1.5 92.1
- OCR-Bench-V2 (EN/ZH) 74.2 / 68.3 74.2
- CC-OCR-Bench-V2 79.6
- MTVQA-Test 56.6
- MADQA 91.8
- QwenVisualOffice 44.6
- RealWorldQA 88
- ERQA 77.8
- LingoQA 84.8
- SURDS 77.8
- SimpleVQA 75
- WorldVQA 53.2
- MMStar 85.9
- PerceptionBench 63.5
- CountQA 82.4
- RefAdv-S 80.2
- Dense200 87
- COCO 78.7
- VisFactor 60.8
- VLMsAreBiased 88.3
- VideoMME (w/ Sub.) 90.4
- VideoMME v2 (w/ Sub.) 68.3
- VideoMMMU 88.7
- MMVU 82.4
- MLVU (M-Avg) 90.8
- TVBench 81.9
- LVBench 81.8
- LVBench (w/ Mem.) 85.6
- EgoLife (w/ Mem.) 80.3
- VideoDR (w/ Search) 73.2
Qwen 3.8 Max
Qwen 3.8 Max ranks first on 24 of the 258 benchmarks with five or more models scored.
279
Benchmarks with a score
24
Ranked first, out of five or more models
52
In the top three
Qwen Aug 2026 Qwen 3.8 Max
Closed weights Flagship
157.6
#29 of 256
Epoch Capabilities Index
Self-Reported Index 160.4 #28 of 404
- MMMU-Pro 82.3% #7 /90
- IFBench 82.8% #2 /80
- DeepSWE 57.5% #16 /28
Where it ranks
Its best score on each benchmark, and where that ranks among every model with a score there. We rank each model once, at its best setting.
Knowledge and language
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| BullshitBench V2 | 95% | #2 of 129 | Official leaderboard | |
| IFBench | 82.8% | #2 of 80 | Lab model card | |
| BullshitBench | 87.3% | #6 of 107 | Official leaderboard | |
| LMArena (Elo) | 1,484 | #20 of 335 | Official leaderboard | |
| LongBench v2 | 66.3% | #3 of 35 | Lab model card | |
| MRCR v2 | 92.9% | #3 of 30 | Lab model card | |
| SimpleVQA | 75% | #4 of 30 | Lab model card | |
| Humanity's Last Exam | 56.2% | #21 of 154 | Lab model card | |
| OpenAI MRCR v2 | 71.2% | #7 of 47 | Official leaderboard | |
| WideSearch | 81.9% | #3 of 19 | Lab model card | |
| ELI5 Arena | 1,656 | #3 of 18 | Official leaderboard | |
| AA-LCR Extended | 73.3% | #9 of 49 | Official leaderboard | |
| Extended NYT Connections | 88.3% | #29 of 103 | Official leaderboard | |
| HLE Extended | 47.1% | #10 of 31 | Official leaderboard | |
| Last Translation Benchmark | 12.5 | #12 of 36 | Official leaderboard | |
| FACTS Multimodal | 51.5% | #8 of 23 | Official leaderboard | |
| FACTS Grounding v2 | 73.8% | #9 of 25 | Official leaderboard | |
| DesignArena ASCII Art | 1,192 | #32 of 77 | Official leaderboard | |
| SimpleQA Verified | 45.8% | #42 of 94 | Epoch AI run | |
| Lech Mazur Creative Writing | 0.12 | #24 of 50 | Official leaderboard | |
| HANDBOOK.md | 30.8% | #8 of 16 | Official leaderboard | |
| Hemingway-bench | 1,009 | #25 of 42 | Official leaderboard | |
| FACTS Benchmark Suite | 60.3% | #16 of 23 | Official leaderboard | |
| Antidote | 985 | #26 of 36 | Official leaderboard | |
| Continual Learning Exam | 26.7% | #9 of 11 | Official leaderboard | |
| FACTS Parametric V2 | 30.8% | #23 of 27 | Official leaderboard | |
| Corp Speak Arena | 1,416 | #19 of 22 | Official leaderboard | |
| Joke Arena | 1,456 | #22 of 25 | Official leaderboard | |
| ComplexConstraints | 46.3% | #14 of 15 | Official leaderboard |
Math and science
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| OTIS Mock AIME 2024-2025 | 99.4% | #18 of 212 | Epoch AI run | |
| GPQA Diamond | 92.7% | #28 of 229 | Epoch AI run | |
| xbench-ScienceQA | 73.4% | #5 of 29 | Official leaderboard | |
| ResearchClawBench | 28.4 | #4 of 23 | Official leaderboard | |
| DTBench | 92% | #38 of 180 | Official leaderboard | |
| HiPhO | 90% | #1 of 4 | Lab model card | |
| ProofBench | 47% | #15 of 59 | Official leaderboard | |
| FrontierMath Tiers 1-3 v2 | 74.7% | #23 of 86 | Official leaderboard | |
| SimpleBench | 62.5% | #33 of 106 | Official leaderboard | |
| FrontierMath Tier 4 v2 | 46.3% | #22 of 68 | Official leaderboard | |
| PhyX | 83.5% | #1 of 3 | Lab model card | |
| Apodex Discovery | 2.66 | #3 of 8 | Official leaderboard | |
| ProofBench v1.1 | 58% | #25 of 52 | Official leaderboard | |
| NatureBench | 11.1% | #9 of 18 | Official leaderboard | |
| FrontierChallenge | 15.5% | #6 of 11 | Official leaderboard | |
| ExplorationBench | 65.7% | #6 of 10 | Official leaderboard | |
| Prophet Arena | 85.6 | #15 of 24 | Official leaderboard | |
| MysteryMechanism | 23.9% | #20 of 31 | Official leaderboard | |
| Riemann-bench | 48% | #14 of 20 | Official leaderboard | |
| PhysClosureBench | 3.46 | #5 of 7 | Official leaderboard | |
| Science's Last Exam | 23.3 | #7 of 9 | Official leaderboard | |
| PaperBenchX | 5.38% | #9 of 10 | Official leaderboard |
Coding
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| IOI (Vals) | 73% | #3 of 55 | Official leaderboard | |
| Data Agent Benchmark | 99.2% | #1 of 18 | Official leaderboard | |
| PaperBench | 93% | #1 of 15 | Lab model card | |
| DesignArena UI Component | 1,336 | #13 of 139 | Official leaderboard | |
| WebDev Arena | 1,671 | #10 of 105 | arena.ai, via Epoch AI | |
| DesignArena Frontend Web App | 1,283 | #6 of 53 | Official leaderboard | |
| Code Arena | 1,670 | #15 of 128 | Official leaderboard | |
| SWE-Bench Pro | 67.7% | #11 of 81 | Lab model card | |
| DesignArena 3D Design | 1,321 | #21 of 136 | Official leaderboard | |
| SWE Atlas Codebase QnA Extended | 33.3% | #6 of 38 | Official leaderboard | |
| Doc2Harness | 45.8% | #3 of 19 | Official leaderboard | |
| DesignArena Game Dev | 1,311 | #23 of 143 | Official leaderboard | |
| DesignArena Full Stack | 1,283 | #8 of 49 | Official leaderboard | |
| QwenBlenderBench | 69.9% | #1 of 6 | Official leaderboard | |
| DesignArena Data Visualization | 1,282 | #21 of 118 | Official leaderboard | |
| APEX-SWE | 52.3% | #12 of 65 | Official leaderboard | |
| AndroidBench | 87% | #6 of 32 | Official leaderboard | |
| DesignArena Code | 1,288 | #28 of 143 | Official leaderboard | |
| Image-to-WebDev Arena | 1,616 | #11 of 55 | Official leaderboard | |
| DesignArena Agentic Overall | 1,252 | #17 of 74 | Official leaderboard | |
| DesignArena Website | 1,270 | #35 of 151 | Official leaderboard | |
| ReactBench | 36.4% | #7 of 28 | Official leaderboard | |
| DesignArena React Native | 1,223 | #15 of 56 | Official leaderboard | |
| Landing Page Arena | 1,566 | #4 of 14 | Official leaderboard | |
| IOI v2 (Vals) | 68.9% | #13 of 45 | Official leaderboard | |
| DesignArena Android (Kotlin) | 1,197 | #16 of 46 | Official leaderboard | |
| SWE Atlas v1.1 | 64.4% | #10 of 28 | Official leaderboard | |
| NL2Repo-Bench | 55.9% | #11 of 30 | Lab model card | |
| QwenReactBench | 1,724 | #2 of 5 | Official leaderboard | |
| QwenSVGBench | 1,713 | #2 of 5 | Official leaderboard | |
| DesignArena Image to HTML | 1,215 | #21 of 51 | Official leaderboard | |
| DesignArena Godot Game Dev | 1,215 | #10 of 24 | Official leaderboard | |
| Vision2Web | 69% | #3 of 7 | Lab model card | |
| SWE-bench Verified Extended | 58.7% | #19 of 44 | Official leaderboard | |
| MirrorCode | 52.2% | #17 of 39 | Epoch AI run | |
| sudo L7 | 16.7% | #8 of 18 | Official leaderboard | |
| Vibe Code Bench | 64.7% | #56 of 120 | Official leaderboard | |
| DAYJOB: Data Science | 7.30% | #7 of 15 | Official leaderboard | |
| TestPrism | 23.7% | #3 of 6 | Official leaderboard | |
| MLS-Bench | 41% | #8 of 15 | Official leaderboard | |
| SWE-Clar | 55% | #5 of 9 | Official leaderboard | |
| DeepSWE | 57.5% | #16 of 28 | Official leaderboard | |
| Vals RSI Index | 0.22 | #16 of 27 | Official leaderboard | |
| QwenQoderBench | 58.4% | #3 of 5 | Official leaderboard | |
| QwenSWEBench | 80.7% | #3 of 5 | Official leaderboard | |
| RSI Bench | 33.4 | #5 of 8 | Official leaderboard | |
| Code Migration | 24% | #49 of 78 | Official leaderboard | |
| SWE-bench CLI | 4.90% | #11 of 16 | Official leaderboard | |
| ScarfBench | 63.7% | #8 of 11 | Official leaderboard | |
| SWE Refactor Bench | 10% | #6 of 8 | Official leaderboard | |
| Android Bench 2.0 | 12.5% | #13 of 17 | Official leaderboard | |
| FrontierCS 2 | 10.4% | #16 of 19 | Official leaderboard | |
| OpenDesign Arena | 72.0 | #18 of 21 | Official leaderboard | |
| FrontierSWE | 15.8% | #22 of 25 | Official leaderboard | |
| Vibe Code Bench 1-100 | 12.8% | #23 of 26 | Official leaderboard | |
| AutoResearchExam | 0.42 | #11 of 12 | Official leaderboard | |
| Snorkel Agentic Coding 2.0 | 14.7% | #15 of 16 | Official leaderboard | |
| RSI-Exam | 32.2% | #17 of 18 | Official leaderboard | |
| FrontierSWE v1 | 73.5% | #5 of 5 | Lab model card |
Agents and tools
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| OSWorld-Verified | 86.1% | #1 of 41 | Lab model card | |
| τ³-bench Banking | 55.1% | #1 of 25 | Official leaderboard | |
| SkillsBench | 70.2% | #1 of 14 | Lab model card | |
| AndroidWorld | 85.3% | #2 of 23 | Lab model card | |
| ScreenSpot-Pro | 84.5% | #5 of 32 | Lab model card | |
| LMArena Search Arena | 1,227 | #6 of 37 | Official leaderboard | |
| Web Context (Nimble) | 77.3% | #2 of 11 | Official leaderboard | |
| Mystery Game Puzzles | 38% | #21 of 84 | Epoch AI run | |
| Terminal-Bench 2.1 | 86.6% | #14 of 56 | Lab model card | |
| WildClawBench | 56.2% | #8 of 31 | Official leaderboard | |
| Chess Puzzles | 46% | #26 of 93 | Epoch AI run | |
| MobileWorld | 77.8% | #1 of 3 | Lab model card | |
| RecreationBench | 51.7% | #1 of 3 | Lab model card | |
| RuneBench | 5.35 | #27 of 76 | Official leaderboard | |
| WebArena-Verified | 66.8% | #3 of 7 | Lab model card | |
| Terminal-Bench 2.1 Extended | 35.4% | #19 of 44 | Official leaderboard | |
| Learn2Play Bench | 72.5 | #5 of 11 | Official leaderboard | |
| o11y-bench | 78.1% | #12 of 22 | Official leaderboard | |
| Agent Arena | 0.11% | #32 of 56 | Official leaderboard | |
| Cresta-Bench | 48.7% | #8 of 14 | Official leaderboard | |
| BrowseComp Extended | 21.5% | #25 of 42 | Official leaderboard | |
| Hermes Bench | 70.5% | #11 of 17 | Official leaderboard | |
| AutomationBench | 27.3% | #36 of 47 | Lab model card | |
| BabaIsBench | 58% | #25 of 32 | Official leaderboard | |
| Long-Horizon-Terminal-Bench 2.0 | 43.5% | #12 of 15 | Official leaderboard | |
| Toolathlon-Verified | 72.5% | #17 of 21 | Lab model card | |
| OSWorld 2.0 | 19.4% | #18 of 19 | Lab model card | |
| TraceTrial-Bench | 70% | #9 of 9 | Official leaderboard | |
| SaaS-Bench | 23.6% | #4 of 4 | Official leaderboard |
Professional work
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| MedXpertQA-MM | 80.4% | #1 of 21 | Lab model card | |
| PMC-VQA | 66.2% | #1 of 10 | Lab model card | |
| SLAKE | 90.8% | #1 of 10 | Lab model card | |
| DesignArena PDF | 1,287 | #4 of 37 | Official leaderboard | |
| TaxEval v2 | 75.5% | #17 of 129 | Official leaderboard | |
| YC-Bench | $2,095,838 | #9 of 59 | Official leaderboard | |
| Legal Research Bench | 47.6% | #16 of 93 | Official leaderboard | |
| Immigration Bench | 70.1% | #5 of 25 | Official leaderboard | |
| DesignArena Python-PPTX Slides | 1,220 | #8 of 36 | Official leaderboard | |
| Harvey LAB (Legal Agent Benchmark) | 10.4% | #18 of 80 | Official leaderboard | |
| CorpFin v2 | 65.8% | #27 of 119 | Official leaderboard | |
| FoodTruck Bench | $44,443 | #14 of 59 | Official leaderboard | |
| E-Commerce Bench | 4.72 | #5 of 19 | Official leaderboard | |
| TRLawBench | 9.44 | #9 of 34 | Official leaderboard | |
| APEX-Agents 1.1 | 74.3% | #19 of 65 | Official leaderboard | |
| APEX-Agents | 63.3% | #18 of 61 | Collected by Epoch AI, source not named | |
| CommerceAgentBench | 52.3% | #4 of 13 | Official leaderboard | |
| Tax Agent Bench | 32.1% | #26 of 84 | Official leaderboard | |
| JobBench | 52.7% | #5 of 16 | Official leaderboard | |
| Public Benefits Bench | 67.1% | #16 of 50 | Official leaderboard | |
| HealthBench | 60.2% | #10 of 31 | Lab model card | |
| MedScribe | 85% | #32 of 97 | Official leaderboard | |
| VWE-Bench | 56.9% | #2 of 6 | Official leaderboard | |
| PresentBench | 79.6% | #1 of 3 | Lab model card | |
| WorkSpaceBench | 67.7% | #1 of 3 | Lab model card | |
| Tuesday Work Index | 54.9 | #16 of 44 | Official leaderboard | |
| OR-License | 78.7% | #4 of 11 | Official leaderboard | |
| LegalBench | 83.6% | #51 of 138 | Official leaderboard | |
| $OneMillion-Bench | 31.9% | #5 of 13 | Official leaderboard | |
| BlenderWorld | 60.6 | #5 of 13 | Official leaderboard | |
| CoWorkBench | 74.8% | #2 of 5 | Official leaderboard | |
| Agents' Last Exam | 27% | #13 of 31 | Official leaderboard | |
| EngiWorld | 26.3% | #3 of 7 | Official leaderboard | |
| DAYJOB: Healthcare | 4.80% | #16 of 37 | Official leaderboard | |
| APEX-1 | 65.3% | #32 of 71 | Official leaderboard | |
| Excel Modeling Benchmark | 60.1% | #35 of 76 | Official leaderboard | |
| DAYJOB: Finance | 8.80% | #19 of 39 | Official leaderboard | |
| GDP.xlsx | 20.3% | #10 of 20 | Official leaderboard | |
| Investor's Last Exam | 62.1 | #5 of 10 | Official leaderboard | |
| Vals Index v2 | 51.8% | #34 of 65 | Official leaderboard | |
| Vals Finance Agent v2 | 50.6% | #43 of 80 | Official leaderboard | |
| MortgageTax | 64% | #48 of 87 | Official leaderboard | |
| APEX-Accounting | 57.2% | #28 of 49 | Official leaderboard | |
| DeepFinance | 30.3% | #14 of 23 | Official leaderboard | |
| MedCode | 40.7% | #59 of 96 | Official leaderboard | |
| DAYJOB: Law | 16.9% | #10 of 16 | Official leaderboard | |
| SnorkelFinance 2.0 | 14.1% | #10 of 16 | Official leaderboard | |
| E-Bench-Code (Tencent) | 67.1% | #5 of 8 | Official leaderboard | |
| GDPval Extended | 0.80% | #27 of 43 | Official leaderboard | |
| RWE-bench | 0% | #10 of 15 | Official leaderboard | |
| Epoch Automation Reports | 53% | #4 of 6 | Official leaderboard | |
| MedXpertQA MM Extended | 51.6% | #28 of 41 | Official leaderboard | |
| Vals Index v2.1 | 48.3% | #35 of 49 | Official leaderboard | |
| SnorkelUnderwrite 2.0 | 20% | #12 of 16 | Official leaderboard | |
| E-Bench (Tencent) | 66.8% | #6 of 8 | Official leaderboard | |
| Excel Modeling Benchmark v1.1 | 66% | #33 of 42 | Official leaderboard | |
| SimuVerity | 25.0 | #4 of 5 | Official leaderboard | |
| SnorkelLegal | 25.6% | #13 of 16 | Official leaderboard | |
| SnorkelManufacturing | 8.30% | #13 of 16 | Official leaderboard | |
| DesignArena Spreadsheet | 1,063 | #29 of 35 | Official leaderboard | |
| GDPx | 3.47% | #17 of 20 | Official leaderboard | |
| Digital Work Index | 11.8 | #14 of 16 | Official leaderboard | |
| PolicyBench | 74.8% | #45 of 50 | Official leaderboard | |
| DAYJOB: Tax | 0% | #14 of 15 | Official leaderboard | |
| SnorkelRevOps | 6.50% | #15 of 16 | Official leaderboard | |
| DAYJOB: Accounting | 4.40% | #16 of 17 | Official leaderboard | |
| PLawBench | 73.2% | #1 of 1 | Lab model card | |
| PRBench | 58.3% | #1 of 1 | Lab model card |
Vision, audio and robotics
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| MATH-Vision (MathVision) | 97.7% | #1 of 51 | Lab model card | |
| Video-MME | 90.4% | #1 of 51 | Lab model card | |
| Vision Arena | 1,299 | #3 of 136 | Official leaderboard | |
| CharXiv | 93.5% | #2 of 75 | Lab model card | |
| MMStar | 85.9% | #1 of 36 | Lab model card | |
| Video-MMMU | 88.7% | #1 of 36 | Lab model card | |
| BabyVision | 91.3% | #1 of 27 | Official leaderboard | |
| MMVU | 82.4% | #1 of 26 | Lab model card | |
| MLVU | 90.8% | #1 of 22 | Lab model card | |
| OCRBench v2 | 74.2% | #1 of 22 | Lab model card | |
| RealWorldQA | 88% | #2 of 38 | Lab model card | |
| ZeroBench | 49% | #5 of 76 | Official leaderboard | |
| LogicVista | 91.9% | #1 of 14 | Lab model card | |
| MMMU-Pro | 82.3% | #7 of 90 | Lab model card | |
| ERQA | 77.8% | #3 of 36 | Lab model card | |
| LingoQA | 84.8% | #1 of 10 | Lab model card | |
| LVBench | 81.8% | #4 of 31 | Lab model card | |
| TVBench | 81.9% | #1 of 7 | Lab model card | |
| MTVQA-Test | 56.6% | #1 of 6 | Lab model card | |
| QwenVisualOffice | 44.6% | #1 of 6 | Official leaderboard | |
| VideoMME v2 (with subtitles) | 68.3% | #1 of 6 | Lab model card | |
| VLMsAreBiased | 88.3% | #1 of 5 | Lab model card | |
| GlobeBench | 25.3% | #4 of 19 | Official leaderboard | |
| VisFactor | 60.8% | #1 of 4 | Lab model card | |
| CharXiv Extended | 82.9% | #10 of 38 | Official leaderboard | |
| BVB | 69.4 | #15 of 45 | Official leaderboard | |
| WorldVQA | 53.2% | #3 of 9 | Lab model card | |
| Claw-Eval-MM | 74.8% | #2 of 6 | Lab model card | |
| CountQA | 82.4% | #1 of 3 | Lab model card | |
| HLE-VL | 52.2% | #1 of 3 | Lab model card | |
| Chartography | 29.1% | #17 of 43 | Official leaderboard | |
| Humanity's Sixth Sense | 33% | #15 of 34 | Official leaderboard | |
| VoxelBench | 1,757 | #26 of 56 | Official leaderboard | |
| MMMU-Pro Extended | 34.8% | #18 of 38 | Official leaderboard | |
| Code2Video Benchmark | 1,533 | #9 of 19 | Official leaderboard | |
| Dense 200 (bbox) | 87% | #1 of 2 | Lab model card | |
| SURDS | 77.8% | #1 of 2 | Lab model card | |
| CineRecon | 45.2 | #8 of 12 | Official leaderboard | |
| GDP.pdf | 42.8% | #13 of 19 | Official leaderboard | |
| Code2Video Arena | 1,475 | #9 of 13 | Official leaderboard | |
| CodeCap-Bench | 48.8 | #5 of 7 | Official leaderboard | |
| MineBench | 1,218 | #54 of 75 | Official leaderboard | |
| InteractiveBench | 22.5% | #9 of 12 | Official leaderboard | |
| Embodied First Exam | 3.73% | #7 of 9 | Official leaderboard | |
| Code2Video-Compass | 12.0 | #12 of 15 | Official leaderboard | |
| PerceptionBench (Moonshot) | 63.5% | #4 of 5 | Lab model card | |
| Frontier3DBench | 46.0 | #10 of 11 | Official leaderboard | |
| CC-OCR-Bench-V2 | 79.6% | #1 of 1 | Lab model card | |
| COCO | 78.7% | #1 of 1 | Lab model card | |
| EgoLife (with memory) | 80.3% | #1 of 1 | Lab model card | |
| LVBench (with memory) | 85.6% | #1 of 1 | Lab model card | |
| MADQA | 91.8% | #1 of 1 | Lab model card | |
| RefAdv-S | 80.2% | #1 of 1 | Lab model card | |
| VideoDR (with search) | 73.2% | #1 of 1 | Lab model card |
Safety and security
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| DistressBench | 77.6% | #7 of 42 | Official leaderboard | |
| SAFE-Teen Health advice | 35% | #4 of 17 | Official leaderboard | |
| SAFE-Teen Urgent safety | 26% | #4 of 17 | Official leaderboard | |
| nocot-bench (NCRI) | 85.4 | #45 of 185 | Official leaderboard | |
| SecIT Bench | 78.5% | #7 of 26 | Official leaderboard | |
| SAFE-Teen Relationship boundaries | 51% | #7 of 17 | Official leaderboard | |
| Last Secure Code Benchmark | 22% | #3 of 6 | Official leaderboard | |
| DeepsecBench | 16.4 | #18 of 29 | Official leaderboard | |
| CyberBench v1.1 (Vals) | 28.6% | #44 of 44 | Official leaderboard |
Other
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| SAGE (Vals) | 51.3% | #14 of 81 | Official leaderboard | |
| Conceptual Reasoning Index | 59.4 | #37 of 151 | Official leaderboard | |
| LMCA | 46.2% | #38 of 147 | Official leaderboard | |
| ACCoRD | 71.4 | #43 of 152 | Official leaderboard | |
| CompassRank | 80.6 | #4 of 14 | Official leaderboard | |
| Vals Multimodal Index | 65.4% | #12 of 33 | Official leaderboard | |
| LiveBench | 60.9% | #21 of 31 | Official leaderboard | |
| Hermes Index | 36.2 | #12 of 17 | Official leaderboard | |
| SAGE v1.1 (Vals) | 81.8% | #25 of 29 | Official leaderboard |
Its lab's numbers put it at 160.4, 2.8 points above its ECI
The Self-Reported Index fits one score to the 41 capability results Qwen reported for this model, on the scale of the Epoch Capabilities Index. Labs choose which benchmarks to show and how to run them, so read it as the lab's claim, not an independent measurement. The 90% interval runs from 158.0 to 162.9. How we compute it.
- MATH-Vision (MathVision) 97.7
- CharXiv 93.5
- PaperBench 93
- MRCR v2 92.9
- GPQA Diamond 92.6
- LogicVista 91.9
- BabyVision 91.3
- SLAKE 90.8
- MLVU 90.8
- Video-MME 90.4
- Video-MMMU 88.7
- RealWorldQA 88
- Terminal-Bench 2.1 86.6
- OSWorld-Verified 86.1
- MMStar 85.9
- AndroidWorld 85.3
- LingoQA 84.8
- ScreenSpot-Pro 84.5
- IFBench 82.8
- MMVU 82.4
- MMMU-Pro 82.3
- WideSearch 81.9
- LVBench 81.8
- MedXpertQA-MM 80.4
- SimpleVQA 75
- OCRBench v2 74.2
- Toolathlon-Verified 72.5
- SkillsBench 70.2
- SWE-Bench Pro 67.7
- LongBench v2 66.3
- PMC-VQA 66.2
- HealthBench 60.2
- DeepSWE 56.6
- Humanity's Last Exam 56.2
- NL2Repo-Bench 55.9
- JobBench 53.4
- WorldVQA 53.2
- ZeroBench 49
- AutomationBench 27.3
- Agents' Last Exam 27
- OSWorld 2.0 19.4
Its lab's numbers next to 15 leaderboards
Where the benchmark's own leaderboard, or Epoch AI's run, also has this model. Gaps mostly come from different harnesses, effort levels and task subsets. Every pair.
| Benchmark | Lab says | Board says | Gap |
|---|---|---|---|
| AndroidBench | 75.1% AndroidBench | 87% Leaderboard | −12 pts |
| DeepSWE | 56.6% DeepSWE 1.1 | 57.5% Leaderboard, xhigh, mini-swe-agent | −0.9 pts |
| JobBench | 53.4% JobBench | 52.7% Leaderboard, OpenCode | +0.7 pts |
| GPQA Diamond | 92.6% GPQA Diamond | 92.7% Epoch AI run, xhigh | Same |
| Agents' Last Exam | 27% Agents' Last Exam (Pass / Score) 27.0 / 52.4 | 27% Leaderboard, xhigh, claude_code | Same |
| BabyVision | 91.3% BabyVision (without CI / with CI) 82.0 / 91.3 | 91.3% Leaderboard, tools | Same |
| CoWorkBench | 74.8% CoWorkBench | 74.8% Leaderboard | Same |
| MLS-Bench | 41% MLS-Bench-Lite | 41% Leaderboard, Claude Code | Same |
| QwenBlenderBench | 69.9% QwenBlenderBench | 69.9% Leaderboard | Same |
| QwenQoderBench | 58.4% QwenQoderBench | 58.4% Leaderboard | Same |
| QwenReactBench | 1,724 QwenReactBench | 1,724 Leaderboard | Same |
| QwenSVGBench | 1,713 QwenSVGBench | 1,713 Leaderboard | Same |
| QwenSWEBench | 80.7% QwenSWEBench | 80.7% Leaderboard | Same |
| QwenVisualOffice | 44.6% QwenVisualOffice | 44.6% Leaderboard | Same |
| ZeroBench | 49% Best of ZeroBench (Pass@5) (without CI / with CI) 24.0 / 49.0 / ZeroBench-Sub | 49% Leaderboard, w/ tools; externally reported: Qwen3.8 release; Code-interpreter tools; 5 samplings. | Same |
Qwen's models over time
Qwen's models in color, against every other model on the Epoch Capabilities Index, by release date.
Source: Epoch Capabilities Index, Epoch AI (CC BY 4.0)
Its launch post
The benchmarks the lab chose to headline, with the scores it gave.