- MMLU-Redux 92.7
- GPQA-Diamond 71.1
- C-Eval 89.6
- LiveBench 2024-11-25 77.1
- IFEval strict prompt 83.4
- Arena-Hard 95.6
- AlignBench v1.1 8.94
- Creative Writing v3 84.6
- WritingBench 80.3
- MATH-500 98
- AIME’24 85.7
- AIME’25 81.5
- ZebraLogic 80.3
- AutoLogi 89
- BFCL v3 70.8
- LiveCodeBench v5 70.7
- CodeForces (Rating / Percentile) 2056
- Multi-IF 71.9
- INCLUDE 78.7
- MMMLU 14 languages 84.3
- MT-AIME2024 80.8
- PolyMath 54.7
- MLogiQA 77.1
- Aider (Pass@2)
Qwen3-235B-A22B
Qwen3-235B-A22B ranks first on 3 of the 99 benchmarks with five or more models scored.
101
Benchmarks with a score
3
Ranked first, out of five or more models
8
In the top three
Qwen Apr 2025 Qwen3-235B-A22B
Open weights Flagship
138.4
#130 of 256
Epoch Capabilities Index
Self-Reported Index 137.6 #199 of 404
- SWE-Bench Pro 21.4% #18 /24
- MultiChallenge 41.2% #35 /42
- GPQA Diamond 70.7% #118 /229
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 | |
|---|---|---|---|---|
| Arena-Hard-Auto | 95.6% | #2 of 39 | Lab model card | |
| AlignBench | 8.94 | #1 of 15 | Lab model card | |
| MLogiQA | 77.1% | #1 of 8 | Lab model card | |
| MT-AIME2024 | 80.8% | #1 of 7 | Lab model card | |
| RULER | 95% | #7 of 42 | Lab model card | |
| Encyclo-K | 43.3% | #5 of 23 | Official leaderboard | |
| Meeseeks Utility Rate | 87.1% | #3 of 10 | Official leaderboard | |
| HABIT-Bench | 33.1% | #2 of 6 | Official leaderboard | |
| Arena-Hard v2.0 | 58.4% | #8 of 23 | Official leaderboard | |
| Fiction.LiveBench | 67.7% | #20 of 57 | Fiction.live leaderboard, via Epoch AI | |
| WritingBench | 74.7 | #17 of 45 | Official leaderboard | |
| LMArena (Elo) | 1,402 | #128 of 335 | Official leaderboard | |
| Wer wird Millionär? LLM Benchmark | 15822.0 | #30 of 77 | Official leaderboard | |
| SuperGPQA | 49.4% | #23 of 59 | Official leaderboard | |
| MMLU-Redux | 92.7% | #18 of 45 | Lab model card | |
| LLM Confabulation Leaderboard | 15.6% | #24 of 58 | Official leaderboard | |
| LongBench v2 | 50.1% | #14 of 33 | Official leaderboard | |
| MMLU-Pro | 68.2% | #91 of 212 | Official leaderboard | |
| C-Eval | 89.6% | #21 of 46 | Lab model card | |
| Vectara Hallucination Leaderboard (2025 dataset) | 9.30% | #48 of 104 | Official leaderboard | |
| INCLUDE | 78.7% | #15 of 31 | Lab model card | |
| MMMLU | 84.3% | #36 of 71 | Lab model card | |
| EQ-Bench 3 | 1,267 | #40 of 78 | Official leaderboard | |
| Multi-IF | 71.9% | #15 of 28 | Lab model card | |
| KINA | 29.4% | #22 of 37 | Official leaderboard | |
| Creative Writing v3 | 63.8 | #74 of 120 | Official leaderboard | |
| IFEval | 83.4% | #92 of 135 | Lab model card | |
| Judgemark v2.1 | 56.2 | #34 of 42 | Official leaderboard | |
| MultiChallenge | 41.2% | #35 of 42 | Official leaderboard | |
| MultiNRC | 17.6% | #32 of 38 | Official leaderboard | |
| ConStory-Bench | 1.45 | #12 of 14 | Official leaderboard | |
| BullshitBench V2 | 6.00% | #122 of 129 | Official leaderboard | |
| BullshitBench | 7.27% | #102 of 107 | Official leaderboard | |
| ComplexNLU (Tencent) | 59.8% | #4 of 4 | Official leaderboard |
Math and science
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| MATH-500 | 98% | #7 of 70 | Lab model card | |
| UGMathBench | 81.8% | #2 of 7 | Official leaderboard | |
| AIME 2024 | 85.7% | #31 of 88 | Lab model card | |
| GPQA Main | 62.3% | #22 of 60 | Official leaderboard | |
| MATH Level 5 | 68.9% | #35 of 94 | Epoch AI run | |
| HiPhO | 57.9% | #15 of 36 | Official leaderboard | |
| Misguided Attention | 49% | #19 of 45 | Official leaderboard | |
| PHYBench | 28.7% | #12 of 28 | Official leaderboard | |
| IneqMath | 6.00% | #15 of 34 | Official leaderboard | |
| ForecastBench | 0.13 | #37 of 82 | Official leaderboard | |
| AutoLogi | 89% | #4 of 8 | Lab model card | |
| GPQA Diamond | 70.7% | #118 of 229 | Epoch AI run | |
| DTBench | 75.7% | #104 of 180 | Official leaderboard | |
| PolyMath | 54.7% | #14 of 24 | Lab model card | |
| SciArena | 1,003 | #20 of 34 | Official leaderboard | |
| AIME 2025 | 81.5% | #86 of 143 | Lab model card | |
| HMMT 2025 | 62.5% | #36 of 53 | Official leaderboard | |
| BRUMO 2025 | 86.7% | #29 of 42 | Official leaderboard | |
| ZebraLogic | 80.3% | #16 of 23 | Lab model card | |
| SimpleBench | 31% | #82 of 106 | Official leaderboard | |
| SMT 2025 | 76.9% | #33 of 40 | Official leaderboard |
Coding
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| LiveCodeBench v5 | 70.7% | #10 of 47 | Lab model card | |
| LiveCodeBench | 87.8% | #5 of 21 | Official leaderboard | |
| AetherCode | 17.6% | #3 of 9 | Official leaderboard | |
| Multi-LCB | 48.9% | #4 of 11 | Official leaderboard | |
| AutoCodeBench | 45.9% | #13 of 34 | Official leaderboard | |
| Codeforces rating | 2,056 | #23 of 47 | Lab model card | |
| LiveCodeBench v6 | 65.9% | #11 of 21 | Official leaderboard | |
| LiveSQLBench | 22.2% | #23 of 36 | Official leaderboard | |
| SWE-Bench Pro | 21.4% | #18 of 24 | Official leaderboard | |
| WeirdML | 37.3% | #88 of 112 | Official leaderboard | |
| DesignArena Website | 1,023 | #128 of 151 | Official leaderboard | |
| ArtifactsBench | 41.1% | #17 of 20 | Official leaderboard | |
| DesignArena Code | 992 | #130 of 143 | Official leaderboard | |
| DesignArena UI Component | 950 | #128 of 139 | Official leaderboard | |
| DesignArena Game Dev | 918 | #137 of 143 | Official leaderboard | |
| DesignArena 3D Design | 841 | #133 of 136 | Official leaderboard | |
| IOI (Vals) | 0% | #55 of 55 | Official leaderboard | |
| SWE-Perf | 0.68% | #2 of 2 | Official leaderboard |
Agents and tools
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Berkeley Function Calling Leaderboard (BFCL) | 70.8% | #14 of 41 | Lab model card | |
| Step Race | 3 | #22 of 59 | Official leaderboard | |
| lmgame-Bench | 26.5 | #12 of 23 | Official leaderboard | |
| ITBench | 46% | #7 of 13 | Official leaderboard | |
| BrowseComp-ZH | 13.2% | #5 of 9 | Official leaderboard | |
| Elimination Game | 4 | #38 of 56 | Official leaderboard | |
| SnakeBench | 20 | #130 of 188 | Official leaderboard | |
| Meta Agents Research Environments (ARE) / Gaia2 | 11.6% | #10 of 13 | Official leaderboard |
Professional work
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| BRIDGE | 48.7 | #12 of 39 | Official leaderboard | |
| LEXam | 47.3% | #14 of 30 | Official leaderboard | |
| LegalBench | 80.2% | #86 of 138 | Official leaderboard | |
| TaxEval v2 | 70.7% | #83 of 129 | Official leaderboard | |
| EnterpriseClawBench | 27% | #9 of 9 | Official leaderboard |
Safety and security
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| BaxBench | 32.6% | #15 of 34 | Official leaderboard | |
| Anthropic Red Team | 99% | #38 of 75 | Official leaderboard | |
| nocot-bench (NCRI) | 70.5 | #96 of 185 | Official leaderboard | |
| SimpleSafetyTests | 98.5% | #44 of 75 | Official leaderboard | |
| HELM Safety | 90.0 | #48 of 75 | Official leaderboard | |
| AIR-Bench 2024 | 56% | #58 of 82 | Official leaderboard | |
| Human Pathogen Capabilities Test | 41.5% | #67 of 86 | Official leaderboard | |
| Molecular Biology Capabilities Test | 41.4% | #71 of 90 | Official leaderboard | |
| FORTRESS | 61.4% | #70 of 73 | Official leaderboard | |
| CyberGym | 1.86% | #33 of 34 | Official leaderboard |
Other
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| LiveBench | 77.1% | #4 of 27 | Lab model card | |
| HELM Capabilities | 72.6 | #17 of 60 | Official leaderboard | |
| LMCA | 25% | #102 of 147 | Official leaderboard | |
| Conceptual Reasoning Index | 36.3 | #107 of 151 | Official leaderboard | |
| ACCoRD | 46.8 | #118 of 152 | Official leaderboard |
Its lab's numbers put it at 137.6, in line with its ECI
The Self-Reported Index fits one score to the 18 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 135.7 to 139.4. How we compute it.
- MATH-500 98
- Arena-Hard-Auto 95.6
- RULER 95
- MMLU-Redux 92.7
- C-Eval 89.6
- AutoLogi 89
- AIME 2024 85.7
- MMMLU 84.3
- IFEval 83.4
- AIME 2025 81.5
- ZebraLogic 80.3
- INCLUDE 78.7
- MLogiQA 77.1
- Multi-IF 71.9
- GPQA Diamond 71.1
- Berkeley Function Calling Leaderboard (BFCL) 70.8
- LiveCodeBench v5 70.7
- PolyMath 54.7
Its lab's numbers next to 10 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 | What differs |
|---|---|---|---|---|
| Creative Writing v3 | 84.6 Creative Writing v3 | 63.8 Leaderboard | +20.8 | |
| WritingBench | 80.3 WritingBench | 74.7 Leaderboard, thinking | +5.57 | Qwen reports an earlier WritingBench evaluation, while the current board uses Claude Sonnet 4.5 to judge responses. |
| MATH-500 | 98% MATH-500 | 94.6% Leaderboard, Vals AI harness | +3.4 pts | |
| Berkeley Function Calling Leaderboard (BFCL) | 70.8% BFCL v3 | 67.9% Leaderboard, native function calling | +2.9 pts | |
| IFEval | 83.4% IFEval strict prompt | 81.6% Leaderboard, strict, share of instructions followed | +1.8 pts | |
| LiveCodeBench v5 | 70.7% LiveCodeBench v5 | 70% Leaderboard | +0.7 pts | |
| AIME 2025 | 81.5% AIME’25 | 80.8% Leaderboard | +0.7 pts | |
| GPQA Diamond | 71.1% GPQA-Diamond | 70.7% Epoch AI run | +0.4 pts | |
| PolyMath | 54.7% PolyMath | 54.6% Leaderboard, thinking | Same | |
| RULER | 95% RULER Avg. (4K to 128K, YaRN x4), non-thinking mode, 128K 90.6. Thinking mode Avg. 92.2 with 8192-token thinking budget (Table 23) | 95% Leaderboard | 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.
Its system card
Every result in the longer documents the lab published with it. 1 of them isn't in the launch post.
Qwen3 Technical Report (arXiv 2505.09388) May 14, 2025
Capabilities
- MMLU-Redux 92.7
- GPQA Diamond 71.1
- C-Eval 89.6
- LiveBench 77.1
- IFEval 83.4
- Arena-Hard-Auto 95.6
- AlignBench 8.94
- Creative Writing v3 84.6
- WritingBench 80.3
- MATH-500 98
- AIME 2024 85.7
- AIME 2025 81.5
- ZebraLogic 80.3
- AutoLogi 89
- Berkeley Function Calling Leaderboard (BFCL) 70.8
- LiveCodeBench v5 70.7
- Codeforces rating 2056
- Multi-IF 71.9
- INCLUDE 78.7
- MMMLU 84.3
- MT-AIME2024 80.8
- PolyMath 54.7
- MLogiQA 77.1
- RULER 95