Qwen 3.5 397B A17B (Thinking): benchmark results
Provider: Alibaba. Released 2026-02-16. Access: Open.
Aggregate Rating 1702 ± 9, rank #355 of 2170 rated models, from 175 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| 3D Scene Structure - Multi-Swap Rearrangement | 93.33 | Correctness (%, human-scored; 300 scenes; correct only with | 100 |
| Nejumi 4 - Toxicity - Prohibited Acts | 100 | Criteria met (%) | 100 |
| Path Planning Optimality Proofs (With Lemmas) | 44.12 | Success rate (%): share of tasks with both the final-answer | 100 |
| Nejumi 4 - BFCL - Live AST | 76.85 | Accuracy (%) | 99 |
| GBQA | 41.13 | Recall (%) in Quality Assurance Mode (the agent also reads t | 95.2 |
| BenchTable - Tech | 85.5 | Weighted Score (%) | 94.8 |
| Video-IFBench | 46.1 | Task-gated instruction satisfaction rate (%; TISR over all f | 94.6 |
| Video-IFBench - Multi | 52.8 | Task-gated instruction satisfaction rate (%; TISR on multi-t | 94.6 |
| Video-IFBench - Nested | 28.2 | Task-gated instruction satisfaction rate (%; TISR on nested | 94.6 |
| Video-IFBench - Selection | 44.9 | Task-gated instruction satisfaction rate (%; TISR on selecti | 94.6 |
| Video-IFBench - Single | 48.5 | Task-gated instruction satisfaction rate (%; TISR on single- | 94.6 |
| BenchTable - Reasoning | 81.7 | Weighted Score (%) | 94.2 |
Interactive version: theaggregate.ai/model?slug=qwen-3-5-397b-a17b-thinking · How It Works · Data refreshed daily, snapshot 2026-11-05.