- SWE-bench Verified (internal scaffold) 77.2
- SWE-bench Pro (Qwen-corrected tasks) 53.5
- SWE-bench Multilingual 71.3
- Terminal-Bench 2.0 (Terminus-2, avg of 5) 59.3
- SkillsBench Avg5 (OpenCode, 78 tasks) 48.2
- QwenWebBench 1487
- NL2Repo 36.2
- Claw-Eval Avg 72.4
- Claw-Eval Pass^3 60.6
- QwenClawBench 53.4
- MMLU-Pro 86.2
- MMLU-Redux 93.5
- SuperGPQA 66
- C-Eval 91.4
- GPQA Diamond 87.8
- HLE 24
- LiveCodeBench v6 83.9
- HMMT Feb 25 93.8
- HMMT Nov 25 90.7
- HMMT Feb 26 84.3
- IMOAnswerBench 80.8
- AIME26 (full AIME 2026 I and II) 94.1
- MMMU 82.9
- MMMU-Pro 75.8
- MathVista mini 87.4
- DynaMath 85.6
- VlmsAreBlind 97
- RealWorldQA 84.1
- MMStar 81.4
- MMBench EN-DEV-v1.1 92.3
- SimpleVQA 56.1
- CharXiv RQ 78.4
- CC-OCR 81.2
- OCRBench 89.4
- ERQA 62.5
- CountBench 97.8
- RefCOCO avg 92.5
- EmbSpatialBench 84.6
- RefSpatialBench 70
- VideoMME(w sub.) 87.7
- VideoMMMU 84.4
- MLVU 86.6
- MVBench 75.5
- V* 94.7
- AndroidWorld 70.3
Qwen3.6 27B
Qwen3.6 27B ranks first on OCRBench v2.
91
Benchmarks with a score
1
Ranked first, out of five or more models
6
In the top three
Qwen Apr 2026 Qwen3.6 27B
Open weights
146.7
#87 of 256
Epoch Capabilities Index
Self-Reported Index 147.8 #112 of 404
- MMMU-Pro 75.8% #37 /90
- NL2Repo-Bench 36.2% #26 /30
- SimpleVQA 56.1% #22 /30
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 | |
|---|---|---|---|---|
| MMLU-Pro | 86.2% | #16 of 167 | Lab model card | |
| SEA-HELM | 74.6 | #16 of 71 | Official leaderboard | |
| MMLU-Redux | 93.5% | #11 of 45 | Lab model card | |
| C-Eval | 91.4% | #13 of 46 | Lab model card | |
| SuperGPQA | 66% | #11 of 35 | Lab model card | |
| Judgemark v4 | 53.4 | #24 of 40 | Official leaderboard | |
| Humanity's Last Exam | 24% | #95 of 154 | Lab model card | |
| SimpleVQA | 56.1% | #22 of 30 | Lab model card | |
| EvoArena Terminal-Bench-Evo | 40.9% | #6 of 8 | Official leaderboard | |
| QIMMA | 44.5% | #20 of 25 | Official leaderboard | |
| EvoArena PersonaMem-Evo | 44.4% | #7 of 8 | Official leaderboard | |
| EQ-Bench 4 | 1,015 | #31 of 32 | Official leaderboard | |
| EvoArena SWE-Chain-Evo | 11.6% | #8 of 8 | Official leaderboard |
Math and science
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| HMMT 2025 | 93.8% | #16 of 54 | Lab model card | |
| OTIS Mock AIME 2024-2025 | 91.1% | #66 of 212 | Epoch AI run | |
| GPQA Diamond | 85.9% | #77 of 229 | Epoch AI run | |
| HMMT Nov 2025 | 90.7% | #13 of 28 | Lab model card | |
| AIME 2026 | 94.1% | #18 of 38 | Lab model card | |
| DTBench | 78.1% | #97 of 180 | Official leaderboard | |
| IMO-AnswerBench | 80.8% | #24 of 37 | Lab model card | |
| FrontierMath Tiers 1-3 v2 | 35.1% | #62 of 86 | Official leaderboard | |
| HMMT 2026 | 84.3% | #14 of 19 | Lab model card | |
| ProofBench | 8.00% | #48 of 59 | Official leaderboard |
Coding
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| SWE-bench Verified | 77.2% | #35 of 155 | Lab model card | |
| LiveCodeBench v6 | 83.9% | #22 of 90 | Lab model card | |
| SWE-bench Multilingual | 71.3% | #43 of 73 | Lab model card | |
| SWE-Bench Pro | 53.5% | #58 of 81 | Lab model card | |
| QwenWebBench | 1,487 | #3 of 4 | Lab model card | |
| Vibe Code Bench | 11.9% | #100 of 120 | Official leaderboard | |
| SWE-rebench | 31.2% | #11 of 13 | Official leaderboard | |
| NL2Repo-Bench | 36.2% | #26 of 30 | Lab model card | |
| AndroidBench | 45.1% | #29 of 32 | Official leaderboard | |
| DAXBench v2 | 59% | #45 of 47 | Official leaderboard |
Agents and tools
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| GameWorld | 43.6% | #7 of 22 | Official leaderboard | |
| Terminal-Bench 2.0 | 59.3% | #21 of 63 | Lab model card | |
| Claw-Eval | 72.4% | #5 of 13 | Lab model card | |
| LLM Chess | 270 | #51 of 123 | Official leaderboard | |
| AndroidWorld | 70.3% | #10 of 23 | Lab model card | |
| ClawProBench | 58.7% | #26 of 49 | Official leaderboard | |
| Claw-Eval Pass^3 | 60.6% | #8 of 13 | Lab model card | |
| WildClawBench | 43.2% | #21 of 31 | Official leaderboard | |
| SkillsBench | 48.2% | #10 of 14 | Lab model card | |
| SnakeBench | 18 | #148 of 188 | Official leaderboard | |
| QwenClawBench | 53.4% | #5 of 6 | Lab model card | |
| o11y-bench | 51.6% | #19 of 22 | Official leaderboard | |
| LiveClawBench | 54.5% | #15 of 17 | Official leaderboard | |
| BabaIsBench | 29% | #29 of 32 | Official leaderboard | |
| Mystery Game Puzzles | 7.00% | #78 of 84 | Epoch AI run | |
| CivBench | 1,391 | #19 of 19 | Official leaderboard | |
| DiG-bench | 1.43% | #7 of 7 | Official leaderboard |
Professional work
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| MortgageTax | 68.3% | #17 of 87 | Official leaderboard | |
| CorpFin v2 | 62.3% | #49 of 119 | Official leaderboard | |
| TaxEval v2 | 71.3% | #74 of 129 | Official leaderboard | |
| Vals Index v1 | 48.7% | #26 of 34 | Official leaderboard | |
| CaseLaw v2 | 53.2% | #41 of 53 | Official leaderboard |
Vision, audio and robotics
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| OCRBench v2 | 68.3% | #1 of 14 | Official leaderboard | |
| MMMU | 82.9% | #10 of 103 | Lab model card | |
| RefCOCO / RefCOCO+ / RefCOCOg | 92.5% | #3 of 29 | Lab model card | |
| CountBench | 97.8% | #3 of 28 | Lab model card | |
| MathVista | 87.4% | #8 of 70 | Lab model card | |
| EmbSpatialBench | 84.6% | #2 of 16 | Lab model card | |
| RefSpatial-Bench | 70% | #3 of 16 | Lab model card | |
| MMStar | 81.4% | #7 of 36 | Lab model card | |
| Video-MME | 87.7% | #10 of 51 | Lab model card | |
| VLMs Are Blind | 97% | #3 of 14 | Lab model card | |
| DynaMath | 85.6% | #5 of 23 | Lab model card | |
| MMBench-EN v1.1 | 92.3% | #6 of 23 | Lab model card | |
| MVBench | 75.5% | #6 of 23 | Lab model card | |
| OCRBench | 89.4% | #9 of 34 | Lab model card | |
| V* | 94.7% | #5 of 17 | Lab model card | |
| MLVU | 86.6% | #7 of 22 | Lab model card | |
| Video-MMMU | 84.4% | #12 of 36 | Lab model card | |
| RealWorldQA | 84.1% | #13 of 38 | Lab model card | |
| MMMU-Pro | 75.8% | #37 of 90 | Lab model card | |
| ERQA | 62.5% | #17 of 36 | Lab model card | |
| CC-OCR | 81.2% | #7 of 14 | Lab model card | |
| CharXiv | 78.4% | #41 of 75 | Lab model card |
Safety and security
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| ABC Bench Screening Evasion | 57.2% | #9 of 17 | Official leaderboard | |
| Molecular Biology Capabilities Test | 47.5% | #51 of 90 | Official leaderboard | |
| ABC Bench Fragment Design | 88.2% | #16 of 27 | Official leaderboard | |
| nocot-bench (NCRI) | 67.4 | #110 of 185 | Official leaderboard | |
| Human Pathogen Capabilities Test v2 (HPCT v2) | 47.3% | #48 of 71 | Official leaderboard | |
| Human Pathogen Capabilities Test | 45.3% | #60 of 86 | Official leaderboard | |
| World-Class Bio | 33.2% | #57 of 74 | Official leaderboard | |
| ABC Bench Liquid-Handling Robot | 45.3% | #30 of 34 | Official leaderboard |
Other
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Political Science LLM Benchmark | 66.6% | #19 of 42 | Official leaderboard | |
| SAGE (Vals) | 45.6% | #39 of 81 | Official leaderboard | |
| LMCA | 34.5% | #80 of 147 | Official leaderboard | |
| Conceptual Reasoning Index | 45.2 | #88 of 151 | Official leaderboard | |
| ACCoRD | 58.9 | #93 of 152 | Official leaderboard | |
| Decision Index | 54.7 | #3 of 3 | Official leaderboard |
Its lab's numbers put it at 147.8, 1.0 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 146.5 to 149.0. How we compute it.
- CountBench 97.8
- VLMs Are Blind 97
- V* 94.7
- AIME 2026 94.1
- HMMT 2025 93.8
- MMLU-Redux 93.5
- RefCOCO / RefCOCO+ / RefCOCOg 92.5
- MMBench-EN v1.1 92.3
- C-Eval 91.4
- HMMT Nov 2025 90.7
- OCRBench 89.4
- GPQA Diamond 87.8
- Video-MME 87.7
- MathVista 87.4
- MLVU 86.6
- MMLU-Pro 86.2
- DynaMath 85.6
- EmbSpatialBench 84.6
- Video-MMMU 84.4
- HMMT 2026 84.3
- RealWorldQA 84.1
- LiveCodeBench v6 83.9
- MMMU 82.9
- MMStar 81.4
- CC-OCR 81.2
- IMO-AnswerBench 80.8
- CharXiv 78.4
- SWE-bench Verified 77.2
- MMMU-Pro 75.8
- MVBench 75.5
- Claw-Eval 72.4
- SWE-bench Multilingual 71.3
- AndroidWorld 70.3
- SuperGPQA 66
- Claw-Eval Pass^3 60.6
- Terminal-Bench 2.0 59.3
- SimpleVQA 56.1
- SWE-Bench Pro 53.5
- SkillsBench 48.2
- NL2Repo-Bench 36.2
- Humanity's Last Exam 24
Its lab's number next to a leaderboard
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 |
|---|---|---|---|
| GPQA Diamond | 87.8% GPQA Diamond | 85.9% Epoch AI run | +1.9 pts |
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.