- Agentic coding DeepSWE 1.1 (best of Claude Code and mini-SWE-agent) 58.7
- Agentic coding SWE-bench Pro (Claude Code, Qwen-corrected tasks) 62.5
- SWE-bench Multilingual (mini-SWE-agent) 81
- Repo-level code generation NL2Repo-Bench (Claude Code) 48.1
- Long-horizon office work CoWorkBench 73.9
- Professional job tasks JobBench 55.7
- Agents' Last Exam Pass@1 24.3 Score 51.2 24.3
- Real-world tool use Toolathlon Verified (Pass@1) 73.5
- Instruction following IFBench 81.3
- Scientific reasoning GPQA Diamond 91.7
- Multidisciplinary reasoning HLE (judged by GPT-4o) 35.9
- Competitive coding LiveCodeBench v6 91.9
- Multimodal tool use ClawEval-MM Pass@3 64.4 Average 60.4 60.4
- Application recreation RecreationBench 49.9
- Mobile use AndroidWorld 84.5
- Computer use OSWorld 2.0 Binary 19.4 Partial 52.3 19.4
- Visual web development Vision2Web 64
- Embodied intelligence ERQA 72.3
- Long video understanding LVBench 76.6
- Real-world perception RealWorldQA 88.5
- Visual math problem solving MathVision Without CI 90.6 With CI 95.7 95.7
- Scientific chart analysis CharXiv (RQ) Without CI 84.6 With CI 90.6 90.6
Qwen3.8 Flash
Qwen3.8 Flash ranks first on RealWorldQA.
41
Benchmarks with a score
1
Ranked first, out of five or more models
3
In the top three
Qwen Aug 2026 Qwen3.8 Flash
Closed weights
152.9
#75 of 404
Self-Reported Index, from the lab's own numbers. No Epoch Capabilities Index score yet.
- IFBench 81.3% #6 /80
- NL2Repo-Bench 48.1% #15 /30
- ERQA 72.3% #4 /36
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 | |
|---|---|---|---|---|
| IFBench | 81.3% | #6 of 80 | Lab model card | |
| AgentIF (2026-10 setting) | 65.6% | #3 of 24 | Official leaderboard | |
| Humanity's Last Exam | 35.9% | #71 of 154 | Lab model card | |
| StudyArena Text | 1,197 | #64 of 85 | Official leaderboard | |
| Antidote | 944 | #33 of 36 | Official leaderboard | |
| Hemingway-bench | 929 | #39 of 42 | Official leaderboard |
Math and science
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| GPQA Diamond | 91.7% | #22 of 302 | Lab model card |
Coding
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| LiveCodeBench v6 | 91.9% | #4 of 90 | Lab model card | |
| SWE-bench Multilingual | 81% | #13 of 73 | Lab model card | |
| SWE-Bench Pro | 62.5% | #23 of 81 | Lab model card | |
| NL2Repo-Bench | 48.1% | #15 of 30 | Lab model card | |
| Vision2Web | 64% | #4 of 7 | Lab model card | |
| FDE-Gym | 0.07 | #8 of 12 | Official leaderboard | |
| DeepSWE | 58.7% | #43 of 56 | Lab model card |
Agents and tools
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| AndroidWorld | 84.5% | #3 of 23 | Lab model card | |
| DuelLab GameBench 2 | 62.2 | #33 of 57 | Official leaderboard | |
| Toolathlon-Verified | 73.5% | #13 of 21 | Lab model card | |
| RecreationBench | 49.9% | #2 of 3 | Lab model card | |
| OSWorld 2.0 | 19.4% | #19 of 19 | Lab model card |
Professional work
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| EnterpriseBench: CoreCraft | 57.4% | #15 of 51 | Official leaderboard | |
| CoWorkBench | 73.9% | #4 of 7 | Lab model card | |
| Tuesday Work Index | 43.9 | #28 of 44 | Official leaderboard | |
| JobBench | 55.7% | #11 of 15 | Lab model card | |
| CorpBench Work | 62.9 | #15 of 17 | Official leaderboard | |
| Agents' Last Exam | 24.3% | #23 of 26 | Lab model card | |
| EngiWorld | 16.7% | #7 of 7 | Official leaderboard |
Vision, audio and robotics
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| RealWorldQA | 88.5% | #1 of 38 | Lab model card | |
| MATH-Vision (MathVision) | 95.7% | #4 of 51 | Lab model card | |
| CharXiv | 90.6% | #6 of 75 | Lab model card | |
| ERQA | 72.3% | #4 of 36 | Lab model card | |
| LVBench | 76.6% | #8 of 31 | Lab model card | |
| Humanity's Sixth Sense | 30.9% | #19 of 34 | Official leaderboard | |
| 4DCodeBench | 0.56 | #9 of 16 | Official leaderboard | |
| Chartography | 17.4% | #25 of 43 | Official leaderboard | |
| ScienceVision | 0.36 | #7 of 12 | Official leaderboard | |
| Claw-Eval-MM | 60.4% | #4 of 6 | Lab model card | |
| CineRecon | 39.3 | #10 of 12 | Official leaderboard | |
| Code2Video-Compass | 12.9 | #13 of 15 | Official leaderboard | |
| WorldAuditBench | 12.7% | #5 of 5 | Official leaderboard |
Safety and security
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| nocot-bench (NCRI) | 74.4 | #78 of 185 | Official leaderboard |
Other
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| SuperCLUE | 39.8 | #13 of 16 | Official leaderboard |
Its lab's numbers put it at 152.9, number 75 of 404
The Self-Reported Index fits one score to the 17 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 150.0 to 155.9. How we compute it.
- MATH-Vision (MathVision) 95.7
- LiveCodeBench v6 91.9
- GPQA Diamond 91.7
- CharXiv 90.6
- RealWorldQA 88.5
- AndroidWorld 84.5
- IFBench 81.3
- SWE-bench Multilingual 81
- LVBench 76.6
- Toolathlon-Verified 73.5
- SWE-Bench Pro 62.5
- DeepSWE 58.7
- JobBench 55.7
- NL2Repo-Bench 48.1
- Humanity's Last Exam 35.9
- Agents' Last Exam 24.3
- OSWorld 2.0 19.4
Its launch post
The benchmarks the lab chose to headline, with the scores it gave.