Alibaba
Qwen3.8 Max (0902)
First Max-tier Qwen with open weights (2.4 trillion parameters). It was hosted first, with the weights about a week later; Epoch AI lists the release as 2 August 2026.
- Announced
- Released
- Access
- Public
Benchmark scores
| Test | Score | Best on the tracker | How it was measured |
|---|---|---|---|
| Epoch Capabilities Indexmany tests combined into one score | 157 points | GPT-6.2 Astra 174 points | Independentlisted as "Qwen 3.8 Max", CI 155.35 to 159.7Run 9 Nov 2026 |
| FrontierSWE V2very long engineering jobs | 18% | Claude Fable 5.2 70% | Independentproximus harness, xhigh effort, mean@5 over 34 tasks, 20-hour budget (best@5 26.8) |
| Terminal-Bench 4.1long command-line tasks | 27% | Claude Opus 5.5 66% | IndependentClaude Code harness, max effort, 330 trials, ±3.87 (95% CI) |
| APEX-Agentsprofessional office tasks | 63% | Gemini 4 Carbon 84% | Independentpass@1, unknown effort; listed as "Qwen3.8-Max" |
| FrontierMath T1–3hard maths problems | 66% | Claude Fable 5.2 95% | Independentxhigh reasoning effort (qwen3.8-max-0902_xhigh)Run 2 Sept 2026 |
| GPQA DiamondPhD-level science questionsPhD experts: 70%Saturated | 92% | Gemini 4 Argon 96% | Independentxhigh reasoning effort (qwen3.8-max-0902_xhigh)Run 2 Sept 2026 |
| FrontierMath T4research-level maths problemsSaturated | 34% | GPT-6.1 Sol 100% | Independentxhigh reasoning effort (qwen3.8-max-0902_xhigh)Run 2 Sept 2026 |
| OTIS Mock AIMEcompetition mathsSaturated | 100% | Top score | Independentxhigh reasoning effort (qwen3.8-max-0902_xhigh)Run 2 Sept 2026 |
Sources
- Qwen/Qwen3.8-2.4T-A95B model cardHugging Face · Dataset · huggingface.co
- AINews: Qwen 3.8 Max (2.4T) and 27BLatent Space · News · latent.space
- OpenRouter models API (created timestamp)OpenRouter · Dataset · openrouter.ai
- Epoch Capabilities IndexEpoch AI · Dataset · epoch.ai
- Epoch AI Benchmarking Hub (model release dates and runs)Epoch AI · Dataset · 9 Nov 2026 · epoch.ai