- Agentic terminal coding, TerminalBench 3.0 29
- Agentic coding, DeepSWE 1.1 69.3
- Repository-level code generation, NL2Repo-Bench 64.9
- Black-box software replication, ProgramBench (Almost Solved) 28
- Machine learning research engineering, MLS-Bench-Lite 50.1
- Repository-level code understanding, SWE-Atlas QnA 66.3
- Long-horizon office work, CoWorkBench 76.1
- Professional job tasks, JobBench 64
- Agentic SaaS workflow, Automation Bench v1.0.6 (Pass@1) 50.8
- Real-world tool use, Toolathlon Verified (Pass@1) 73.3
- STEM visual reasoning, MMMU-Pro 82.7
- Embodied intelligence, ERQA 78.3
- Multimodal tool use, ClawEval-MM (Pass@3) 80.2
- General visual reasoning, BabyVision (w/ CI) 93.8
Qwen3.8 Max (0902)
Qwen3.8 Max (0902) ranks first on 3 of the 50 benchmarks with five or more models scored.
50
Benchmarks with a score
3
Ranked first, out of five or more models
9
In the top three
Qwen Sep 2026 Qwen3.8 Max (0902)
Closed weights
156.0
#40 of 256
Epoch Capabilities Index
Self-Reported Index 166.2 #12 of 404
- MMMU-Pro 82.7% #6 /90
- NL2Repo-Bench 64.9% #3 /30
- Terminal-Bench 4.0 27% #20 /40
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 | |
|---|---|---|---|---|
| SimpleQA Verified | 47.3% | #37 of 94 | Epoch AI run | |
| StudyArena Text | 1,212 | #60 of 85 | Official leaderboard |
Math and science
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| OTIS Mock AIME 2024-2025 | 100% | #6 of 212 | Epoch AI run | |
| GPQA Diamond | 92.3% | #30 of 229 | Epoch AI run | |
| NatureBench | 17.8% | #3 of 18 | Official leaderboard | |
| FrontierMath Tiers 1-3 v2 | 65.6% | #37 of 86 | Official leaderboard | |
| FrontierMath Tier 4 v2 | 34.1% | #30 of 68 | Official leaderboard |
Coding
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| NL2Repo-Bench | 64.9% | #3 of 30 | Lab model card | |
| Code Arena | 1,676 | #13 of 128 | Official leaderboard | |
| SWE Atlas | 66.3% | #2 of 14 | Lab model card | |
| Image-to-WebDev Arena | 1,638 | #8 of 55 | Official leaderboard | |
| ALE-Bench | 1,964 | #24 of 126 | Official leaderboard | |
| MLS-Bench | 50.1% | #1 of 5 | Lab model card | |
| Vibe Code Bench 1-100 | 24.8% | #7 of 26 | Official leaderboard | |
| Vibe Code Bench | 77.1% | #40 of 120 | Official leaderboard | |
| Code Migration v1.1 | 57.2% | #11 of 30 | Official leaderboard | |
| DeepSWE | 69.3% | #21 of 56 | Lab model card | |
| DAXBench v2 | 81.7% | #20 of 47 | Official leaderboard | |
| RSI-Exam | 39.2% | #8 of 18 | Official leaderboard | |
| MirrorCode | 47.8% | #19 of 39 | Epoch AI run | |
| ProgramBench almost solved | 28% | #5 of 8 | Lab model card | |
| FrontierSWE | 17.8% | #21 of 25 | Official leaderboard |
Agents and tools
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| AutomationBench | 50.8% | #13 of 47 | Lab model card | |
| DuelLab GameBench 2 | 69.6 | #22 of 57 | Official leaderboard | |
| Chess Puzzles | 37% | #44 of 93 | Epoch AI run | |
| Terminal-Bench 4.0 | 27% | #20 of 40 | Official leaderboard | |
| RecreationBench | 34.8% | #5 of 10 | Official leaderboard | |
| Terminal-Bench 3.0 | 29% | #3 of 6 | Lab model card | |
| Terminal-Bench 4.1 | 27% | #22 of 43 | Official leaderboard | |
| Toolathlon-Verified | 73.3% | #14 of 21 | Lab model card |
Professional work
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Tax Agent Bench | 49.7% | #4 of 84 | Official leaderboard | |
| Excel Modeling Benchmark | 72% | #9 of 76 | Official leaderboard | |
| CoWorkBench | 76.1% | #1 of 7 | Lab model card | |
| Vals Finance Agent v2 | 57.9% | #12 of 80 | Official leaderboard | |
| Vals Legal Index | 73.9% | #7 of 38 | Official leaderboard | |
| Vals Finance Index | 61.7% | #8 of 40 | Official leaderboard | |
| Legal Research Bench | 46.1% | #21 of 93 | Official leaderboard | |
| Harvey LAB v1.1 (HLAB-Vals-1.1) | 15.4% | #11 of 38 | Official leaderboard | |
| Excel Modeling Benchmark v1.1 | 72% | #14 of 42 | Official leaderboard | |
| JobBench | 64% | #6 of 15 | Lab model card | |
| Vals Index v2.2 | 59% | #17 of 37 | Official leaderboard | |
| Vals Finance Agent v2.1 | 58.8% | #30 of 40 | Official leaderboard |
Vision, audio and robotics
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| ERQA | 78.3% | #2 of 36 | Lab model card | |
| MMMU-Pro | 82.7% | #6 of 90 | Lab model card | |
| BabyVision | 93.8% | #2 of 26 | Lab model card | |
| Claw-Eval-MM | 80.2% | #1 of 6 | Lab model card | |
| ScienceVision | 0.38 | #6 of 12 | Official leaderboard | |
| Furniture Assembly Benchmark | 20% | #35 of 35 | Official leaderboard |
Safety and security
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| CWE-bench 2 | 45.4% | #12 of 14 | Official leaderboard |
Other
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| SuperCLUE | 43.2 | #8 of 16 | Official leaderboard |
Its lab's numbers put it at 166.2, 10.3 points above its ECI
The Self-Reported Index fits one score to the 8 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 162.6 to 169.9. How we compute it.
- BabyVision 93.8
- MMMU-Pro 82.7
- Toolathlon-Verified 73.3
- DeepSWE 69.3
- SWE Atlas 66.3
- NL2Repo-Bench 64.9
- JobBench 64
- AutomationBench 50.8
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.