Qwen 3.8 Max (0902): benchmark results
Alibaba's September 2026 checkpoint refresh of Qwen3.8-Max, a 2.4-trillion-parameter, 95-billion-active Mixture-of-Experts model further post-trained for coding and enterprise work, with the same 1-million-token context. Provider: Alibaba. Released 2026-09-02. Access: API.
Aggregate Rating 1841 ± 10, rank #36 of 1635 rated models, from 179 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Doc2Harness - Interactive terminal | 93.3 | Pass rate (%) | 100 |
| SEAL - SWE Atlas - Codebase QnA | 66.3 | Score | 100 |
| SuperCLUE General (July 2026) - Overall | 72.62 | Score | 100 |
| World Agent | 81.3 | Maintenance completion score (%): share of a world's contrac | 100 |
| World Agent - Relation | 73.8 | Maintenance relation score (%): share of authored causal, te | 100 |
| OTIS Mock AIME 2024-25 | 100 | Accuracy (%) | 98.1 |
| NYT Connections Older Models | 88.8 | Score (%) | 97.3 |
| Tau3 Banking | 47.84 | Success Rate (%) | 97.1 |
| AA APEX-Agents | 42.4 | Pass@1 (%) | 96.9 |
| EnterpriseOps-Gym-AA - Calendar | 50.96 | Task Success (%) | 96.9 |
| LMArena WebDev Arena | 1687.74 | Arena rating (imported) | 96.8 |
| AA Terminal-Bench 2.1 | 88.76 | Pass@1 (%) | 96.7 |
Interactive version: theaggregate.ai/model?slug=qwen-3-8-max-0902 · How It Works · Data refreshed daily, snapshot 2026-11-05.