Qwen3-235B-A22B-Thinking ranks first on 2 of the 76 benchmarks with five or more models scored.
Qwen Jul 2025 Jul 2025 snapshot Open weights Flagship
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.
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Encyclo-K | 51.8% | #1 of 23 | Official leaderboard | |
| WritingBench | 82.3 | #2 of 45 | Official leaderboard | |
| Creative Writing v3 | 86.1 | #1 of 12 | Lab model card | |
| LongBench v2 | 60.6% | #3 of 33 | Official leaderboard | |
| Multi-IF | 80.6% | #4 of 28 | Lab model card | |
| MMLU-Redux | 93.8% | #8 of 45 | Lab model card | |
| Arena-Hard v2.0 | 79.7% | #5 of 24 | Lab model card | |
| MMLU-Pro | 84.4% | #38 of 167 | Lab model card | |
| SimpleQA Verified | 53.7% | #14 of 56 | Official leaderboard | |
| Fiction.LiveBench | 75% | #15 of 57 | Fiction.live leaderboard, via Epoch AI | |
| MMLU-ProX | 81% | #8 of 23 | Lab model card | |
| INCLUDE | 81% | #11 of 31 | Lab model card | |
| SuperGPQA | 64.9% | #13 of 35 | Lab model card | |
| LMArena (Elo) | 1,400 | #132 of 335 | Official leaderboard | |
| IFEval | 87.8% | #57 of 135 | Lab model card | |
| KINA | 32.1% | #17 of 37 | Official leaderboard | |
| LLM Confabulation Leaderboard | 16.8% | #29 of 58 | Official leaderboard | |
| ConStory-Bench | 0.56 | #7 of 14 | Official leaderboard | |
| MultiNRC | 27.1% | #22 of 38 | Official leaderboard | |
| RULERv2 | 85.2% | #2 of 3 | Official leaderboard | |
| Humanity's Last Exam | 18.2% | #112 of 154 | Lab model card |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| AMO-Bench | 47.8% | #2 of 21 | Official leaderboard | |
| PHYBench | 35% | #6 of 28 | Official leaderboard | |
| AIME 2025 | 92.3% | #39 of 143 | Lab model card | |
| HiPhO | 61.5% | #10 of 36 | Official leaderboard | |
| SciArena | 1,036 | #11 of 34 | Official leaderboard | |
| OTIS Mock AIME 2024-2025 | 86.7% | #78 of 212 | Epoch AI run | |
| PolyMath | 60.1% | #10 of 24 | Lab model card | |
| GPQA Diamond | 80% | #96 of 229 | Epoch AI run | |
| DTBench | 80.3% | #90 of 180 | Official leaderboard | |
| MathArena (Apex, IMO, USAMO) | 5.21% | #25 of 47 | Official leaderboard | |
| FrontierMath | 8.48% | #33 of 55 | Official leaderboard | |
| HMMT 2025 | 83.9% | #33 of 54 | Lab model card | |
| ForecastBench | 0.14 | #58 of 82 | Official leaderboard | |
| General365 | 47.9% | #20 of 23 | Official leaderboard | |
| FrontierMath Tier 4 | 0% | #51 of 55 | Epoch AI run |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| ArtifactsBench | 55% | #6 of 20 | Official leaderboard | |
| BigCodeArena | 1,047 | #4 of 13 | Official leaderboard | |
| Codeforces rating | 2,134 | #15 of 47 | Lab model card | |
| AutoCodeBench | 47.7% | #11 of 34 | Official leaderboard | |
| LiveCodeBench v6 | 74.1% | #48 of 90 | Lab model card | |
| OJBench | 32.5% | #10 of 16 | Lab model card | |
| WeirdML | 41% | #73 of 112 | Official leaderboard | |
| DesignArena Website | 1,046 | #125 of 151 | Official leaderboard | |
| DesignArena Code | 1,023 | #123 of 143 | Official leaderboard | |
| DesignArena 3D Design | 980 | #122 of 136 | Official leaderboard | |
| DesignArena UI Component | 936 | #130 of 139 | Official leaderboard | |
| DesignArena Game Dev | 947 | #134 of 143 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Berkeley Function Calling Leaderboard (BFCL) | 71.9% | #11 of 41 | Lab model card | |
| LisanBench | 803.7 | #60 of 133 | Official leaderboard | |
| Meta Agents Research Environments (ARE) / Gaia2 | 15.7% | #8 of 13 | Official leaderboard | |
| Pencil Puzzle Bench | 0% | #35 of 53 | Official leaderboard | |
| Galileo Agent Leaderboard v2 | 34% | #15 of 21 | Official leaderboard | |
| VITA-Bench | 14.5% | #13 of 18 | Official leaderboard | |
| tau-bench | 56.9% | #18 of 23 | Lab model card | |
| Mystery Game Puzzles | 9.00% | #71 of 84 | Epoch AI run | |
| tau2-bench | 58.5% | #67 of 79 | Lab model card | |
| SnakeBench | 17 | #162 of 188 | Official leaderboard | |
| AgencyBench | 27% | #9 of 9 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| BizFinBench v2 | 53.3% | #2 of 9 | Official leaderboard | |
| StockBench | 4.00 | #4 of 13 | Official leaderboard | |
| BRIDGE | 48.1 | #15 of 39 | Official leaderboard | |
| PLawBench | 62.8% | #5 of 11 | Official leaderboard | |
| ProfBench | 41.1% | #62 of 72 | Official leaderboard | |
| Vending-Bench 2 | $-11 | #64 of 67 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| BioTIER Permit | 99.7% | #20 of 69 | Official leaderboard | |
| nocot-bench (NCRI) | 68.9 | #105 of 185 | Official leaderboard | |
| Human Pathogen Capabilities Test | 49% | #50 of 86 | Official leaderboard | |
| SUPERChem | 29.4% | #4 of 6 | Official leaderboard | |
| Molecular Biology Capabilities Test | 44.4% | #62 of 90 | Official leaderboard | |
| Agentic BAIM-LLM Evaluation (ABLE) | 45.2% | #13 of 18 | Official leaderboard | |
| Human Pathogen Capabilities Test v2 (HPCT v2) | 43.1% | #54 of 71 | Official leaderboard | |
| BioTIER Refuse | 20.9% | #59 of 69 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| LiveBench | 78.4% | #3 of 27 | Lab model card | |
| Conceptual Reasoning Index | 42.4 | #92 of 151 | Official leaderboard | |
| LMCA | 29.3% | #93 of 147 | Official leaderboard | |
| ACCoRD | 57.1 | #98 of 152 | Official leaderboard |
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 138.1 to 142.4. How we compute it.
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 |
|---|---|---|---|---|
| WritingBench | 88.3 WritingBench | 82.3 Leaderboard | +5.96 | The lab published its score before WritingBench switched judges, while the current board uses Claude Sonnet 4.5. |
| GPQA Diamond | 81.1% GPQA | 80% Epoch AI run | +1.0 pts | |
| MMLU-Pro | 84.4% MMLU-Pro | 84.5% Leaderboard | Same |
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)
The benchmarks the lab chose to headline, with the scores it gave.