Qwen3-Coder-480B-A35B ranks first on Multi-SWE-bench mini.
Qwen Jul 2025 Open weights
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 | |
|---|---|---|---|---|
| LMArena (Elo) | 1,388 | #146 of 335 | Official leaderboard | |
| BullshitBench | 20% | #69 of 107 | Official leaderboard | |
| BullshitBench V2 | 20% | #94 of 129 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Multi-SWE-bench mini | 25.8% | #1 of 5 | Official leaderboard | |
| Bird-SQL | 68.1% | #3 of 13 | Official leaderboard | |
| SWE-bench Verified | 69.6% | #16 of 60 | Official leaderboard | |
| LiveSQLBench | 26.8% | #14 of 36 | Official leaderboard | |
| Aider Polyglot | 61.8% | #14 of 30 | Lab model card | |
| AutoCodeBench | 44.8% | #16 of 34 | Official leaderboard | |
| SWE-Bench Pro | 38.7% | #13 of 24 | Official leaderboard | |
| ABC-Bench | 43.1% | #4 of 7 | Official leaderboard | |
| SWE-bench-Live | 24.7% | #7 of 12 | Official leaderboard | |
| Spider 2.0 | 31.1% | #14 of 23 | Official leaderboard | |
| WeirdML | 41.2% | #72 of 112 | Official leaderboard | |
| ArtifactsBench | 47.1% | #14 of 20 | Official leaderboard | |
| AlgoTune | 1.44× | #13 of 18 | Official leaderboard | |
| GSO-Bench | 4.90% | #23 of 30 | Official leaderboard | |
| Code Arena | 1,275 | #111 of 128 | Official leaderboard | |
| WebDev Arena | 1,274 | #92 of 105 | arena.ai, via Epoch AI | |
| SWE-bench Multilingual | 54.7% | #66 of 73 | Lab model card | |
| Snorkel Agentic Coding | 18.8% | #10 of 11 | Official leaderboard | |
| ALE-Bench | 822 | #122 of 126 | Official leaderboard | |
| OpenHands Index | 30.9% | #33 of 33 | Official leaderboard | |
| AetherCode | 7.70% | #9 of 9 | Official leaderboard | |
| CodeClash | 952 | #8 of 8 | Official leaderboard | |
| Multi-SWE-bench flash | 27% | #1 of 1 | Lab model card |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| tau-bench | 68.8% | #6 of 23 | Lab model card | |
| MCPMark | 24.8% | #12 of 28 | Official leaderboard | |
| Berkeley Function Calling Leaderboard (BFCL) | 68.7% | #22 of 41 | Lab model card | |
| Terminal-Bench | 27.2% | #11 of 18 | tbench.ai, via Epoch AI | |
| SnakeBench | 21 | #116 of 188 | Official leaderboard | |
| MCP-Universe | 22.9% | #19 of 30 | Official leaderboard | |
| LisanBench | 213.7 | #87 of 133 | Official leaderboard | |
| Pencil Puzzle Bench | 0% | #36 of 53 | Official leaderboard | |
| Terminal-Bench 2.0 | 27.2% | #35 of 43 | Official leaderboard | |
| Toolathlon | 14.5% | #45 of 50 | Official leaderboard | |
| WebArena | 49.9% | #8 of 8 | Lab model card | |
| Mind2Web | 55.8% | #1 of 1 | Lab model card |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Physical-Engineering's Last Exam | 0.06 | #12 of 18 | Official leaderboard | |
| StockBench | 10.0 | #10 of 13 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| BaxBench | 37.2% | #11 of 34 | Official leaderboard | |
| nocot-bench (NCRI) | 73.1 | #83 of 185 | Official leaderboard | |
| AutoBaxBench | 16.4% | #7 of 11 | Official leaderboard |
The Self-Reported Index fits one score to the 6 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 136.9 to 143.2. 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 |
|---|---|---|---|
| Multi-SWE-bench flash | 27% Multi-SWE-bench flash | 25.3% Leaderboard, unverified, MopenHands | +1.7 pts |
| SWE-bench-Live | 26.3% SWE-bench Live | 24.7% Leaderboard, OpenHands | +1.6 pts |
| Multi-SWE-bench mini | 25.8% Multi-SWE-bench mini | 25.8% Leaderboard, self-reported, MopenHands | Same |
| Spider 2.0 | 31.1% Spider2 | 31.1% Leaderboard, Spider-Agent | Same |
| SWE-bench Verified | 69.6% SWE-bench Verified w/ OpenHands, 500 turns | 69.6% Leaderboard, OpenHands | Same |
The benchmarks the lab chose to headline, with the scores it gave.