Qwen Max: benchmark results
Alibaba's flagship Qwen Max model, represented here as a 700B-parameter closed model. Provider: Alibaba. Released 2025-04-28. Access: API.
Unified ELO 1628 ± 13, rank #355 of 1639 rated models, from 70 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| pfgen-bench - QA Mode - Helpfulness | 0.68 | Helpfulness Score | 97.6 |
| pfgen-bench - QA Mode - Score | 0.84 | pfgen Score (mean of three) | 97 |
| pfgen-bench - QA Mode - Fluency | 0.86 | Fluency Score | 95.2 |
| pfgen-bench - QA Mode - Truthfulness | 0.96 | Truthfulness Score | 95.2 |
| OpenCompass LLM - Average - English (CompassBench 2402) | 56.3 | Score (%) | 94.1 |
| OpenCompass LLM - Agent (CompassBench 2402) | 66 | Score (%) | 91.2 |
| OpenCompass LLM - Agent - Chinese (CompassBench 2402) | 62.2 | Score (%) | 91.2 |
| OpenCompass LLM - Agent - English (CompassBench 2402) | 69.7 | Score (%) | 91.2 |
| CSimpleQA | 57.4 | Score (%) | 90 |
| Plan-RewardBench - Multi-Turn Planning (Hard) | 66.71 | Pairwise accuracy (%) on the 73 hard multi-turn planning pai | 90 |
| ACEBench | 81.7 | Overall (self-reported) | 89.3 |
| SuiChat-CN - Few-Shot | 73.6 | Macro F1 (%) of suicide-risk level classification of Chinese | 88.4 |
Interactive version: theaggregate.ai/model?slug=qwen-max · How It Works · Data refreshed daily, snapshot 2026-10-09.