Qwen 3.5 397B A17B: benchmark results
Alibaba's open Qwen 3.5 397B A17B, a sparse MoE with 17B active parameters (February 2026). Provider: Alibaba. Released 2026-02-16. Access: Open.
Aggregate Rating 1731 ± 3, rank #146 of 1656 rated models, from 1975 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 |
| AgentCE-Bench | 84.9 | Task score (%) on AgentCE-Bench over all 324 instances, the | 100 |
| AgentCE-Bench - Course Scheduling | 94.4 | Task score (%) on AgentCE-Bench course domain (54 instances) | 100 |
| AgentCE-Bench - Grocery Shopping | 83.3 | Task score (%) on AgentCE-Bench shopping domain (54 instance | 100 |
| AgentCE-Bench - Meal Planning | 81.5 | Task score (%) on AgentCE-Bench meal domain (54 instances): | 100 |
| AgentCE-Bench - PC Build Configuration | 81.5 | Task score (%) on AgentCE-Bench pc_build domain (54 instance | 100 |
| AgentCE-Bench - Travel Itinerary | 77.8 | Task score (%) on AgentCE-Bench travel domain (54 instances) | 100 |
| AgentCE-Bench - Workforce Scheduling | 90.7 | Task score (%) on AgentCE-Bench workforce domain (54 instanc | 100 |
| AirGroundBench - Viewpoint Reasoning | 48.35 | Accuracy (%) on the viewpoint spatial-reasoning VQA task; hi | 100 |
| BEAR-Bench | 75.4 | Overall (%) | 100 |
| Chronicles-OCR - Mature Text Parsing | 0.73 | Paragraph-level normalized edit similarity (0-1: one minus t | 100 |
| DORA - Multimodal Report Synthesis | 49.17 | Final-answer accuracy (%; ReAct agent with tools) | 100 |
Interactive version: theaggregate.ai/model?slug=qwen-3-5-397b-a17b · How It Works · Data refreshed daily, snapshot 2026-10-20.