Qwen 3.8 Max: benchmark results
Alibaba's Qwen3.8-Max (August 3, 2026), the largest Qwen model at release, a 2.4-trillion-parameter, 95-billion-active MoE with text, image and video input. Provider: Alibaba. Released 2026-08-03. Access: API.
Aggregate Rating 1843 ± 4, rank #34 of 1635 rated models, from 809 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| BabelArena - DeepPlanning | 46 | Pass^1 (%; DeepPlanning (long-horizon travel and shopping pl | 100 |
| BabelArena - VitaBench | 42.3 | Pass^1 (%; VitaBench (food delivery, in-store consumption an | 100 |
| CLE - Hermes-Agent Trial 1 Completion | 25.67 | Completion (%) | 100 |
| CLE - Hermes-Agent Trial 1 Level 1 | 70.83 | Cumulative pass rate (%) | 100 |
| CLE - Hermes-Agent Trial 2 Level 1 | 74.17 | Cumulative pass rate (%) | 100 |
| CLE - OpenClaw Trial 1 Level 1 | 60 | Cumulative pass rate (%) | 100 |
| CLE - OpenClaw Trial 2 Level 1 | 65 | Cumulative pass rate (%) | 100 |
| DeliverBench - Process evaluation - P03 | 84 | Rubric score (0–100) | 100 |
| DoGBench | 47.26 | Composite score (0-100), held-out split (117 items) | 100 |
| GMA | 67.33 | Success rate (%; a task succeeds only when all its subtasks | 100 |
| GMA - Atomic | 93.15 | Success rate (%; 300 tasks in seven custom Android applicati | 100 |
| GMA - Compositional | 70.54 | Success rate (%; 300 tasks in seven custom Android applicati | 100 |
Interactive version: theaggregate.ai/model?slug=qwen-3-8-max · How It Works · Data refreshed daily, snapshot 2026-11-05.