Qwen 3 Max (Preview): benchmark results
Alibaba's preview release of Qwen 3 Max. Provider: Alibaba. Released 2025-09-05. Access: API.
Aggregate Rating 1686 ± 10, rank #230 of 1656 rated models, from 303 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Nejumi 4 - jaster (2-shot) - JSeM | 83 | Exact match (%) | 96.5 |
| ALPBench - Fishing Rods (Profile F1) | 18.33 | Profile-level F1 (%): the harmonic mean of the shares of pre | 96.4 |
| KAware - Internal Function Pass Rate | 95.98 | The end-to-end pass rate (%): share of tasks whose trajector | 94.1 |
| ALPBench - Badminton Rackets (Profile F1) | 23.67 | Profile-level F1 (%): the harmonic mean of the shares of pre | 92.9 |
| KAware - Hybrid Composition Pass Rate | 97.01 | The end-to-end pass rate (%): share of tasks whose trajector | 91.2 |
| Nejumi 4 - JBBQ (2-shot) - Accuracy | 94 | Accuracy (%) | 90.2 |
| Nejumi 4 - BFCL - Non-Live AST | 81.11 | Accuracy (%) | 89.9 |
| ReLE - Finance | 85.1 | Accuracy (%) | 89.5 |
| AA Omniscience - Software Engineering (SWE) - R | 30 | Accuracy (%) | 88.7 |
| KAware - External Function | 97.36 | The knowing-acting score (%, KAS): harmonic mean of knowing | 88.2 |
| Nejumi 4 - HalluLens | 99 | Hallucination resistance (%) | 88.1 |
| AA Omniscience - Software Engineering (SWE) - C | 56 | Accuracy (%) | 87.5 |
Interactive version: theaggregate.ai/model?slug=qwen-3-max-preview · How It Works · Data refreshed daily, snapshot 2026-10-20.