Qwen 3.7 Max: benchmark results
Alibaba's flagship Qwen 3.7 model for broad high-end chat and reasoning workloads. Provider: Alibaba. Released 2026-05-20. Access: API.
Aggregate Rating 1786 ± 4, rank #75 of 1656 rated models, from 726 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| ATOBench - Basket (Native) | 66.7 | Grounded verification rate (%; the resource-ownership contra | 100 |
| AeroCopilotBench - Tier-1 Aviation Knowledge | 89.17 | Accuracy (%; 1,200 multiple-choice questions) | 100 |
| AppWorld-UL (Code Agent) - Confirmation Tasks | 67.7 | Interactive task goal completion (%; share of 102 tasks whos | 100 |
| AppWorld-UL (Function Calling) - Confirmation Tasks | 43.1 | Interactive task goal completion (%; share of 102 tasks whos | 100 |
| AppWorld-UL (Function Calling) - Scenario Goal Completion | 13.9 | Interactive scenario goal completion (%; share of task scena | 100 |
| Arcee MMLU-Pro (BenchLM) | 89.6 | Accuracy (%) | 100 |
| CompanionBench - English - Holding Ambiguity | 8.75 | Mean judge score (1-10) | 100 |
| DynamicMCPBench | 51.2 | Pass^3 (%; share of the 750 balanced tasks from 15 tool-use | 100 |
| FinRiskAtlas - FinRisk-Ask - Conditional Request Alignment | 87.67 | CRA (%; DeepSeek-V4-Flash request-alignment judge) | 100 |
| HMMT February 2026 | 97.1 | Score (mixed sources) | 100 |
| HoosierHelp - Resource Exact Match | 67.5 | Resource exact match (%; conversations ending in a final_rec | 100 |
| Horangi 4 - KorQuAD 1.0 | 90.34 | F1 (x100) | 100 |
Interactive version: theaggregate.ai/model?slug=qwen-3-7-max · How It Works · Data refreshed daily, snapshot 2026-10-20.