Qwen 3.5 35B A3B: benchmark results
Alibaba's open Qwen 3.5 35B A3B MoE (3B active). Provider: Alibaba. Released 2026-02-24. Access: Open.
Aggregate Rating 1676 ± 4, rank #256 of 1656 rated models, from 1668 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| AgentCollabBench | 0.9 | Instruction Decay Rate (%) | 100 |
| AgentCollabBench - RTD | 94 | Radioactive Tracer Durability (%) | 100 |
| DeepTumorVQA | 39.9 | Overall Accuracy (%) | 100 |
| DistortBench - Blur | 68.3 | Joint accuracy (%) on the three blur types (Gaussian, lens, | 100 |
| FinInteract | 28.9 | Accuracy (%; all 173 bilingual instances; ReAct agent in the | 100 |
| GRACE - 2WikiMHQA | 83.08 | Step-level faithful/unfaithful F1 (%) on 2WikiMHQA grounding | 100 |
| HoloCount - Partial Object Occlusion | 65 | Accuracy (%; exact-match accuracy of the single integer coun | 100 |
| Hypersim | 13.1 | Score (score_points) | 100 |
| LLM Stats (AndroidWorld_SR) | 71.1 | Score (%) | 100 |
| LinkS2Bench - Relative Direction | 38.3 | Relative Direction: multiple-choice accuracy (%) on the dire | 100 |
| MemTrace - Conflict Resolution | 88.6 | Gist accuracy (%; GPT-4o judge; the model answers from the v | 100 |
| MemTrace - Fresh | 61.3 | Gist accuracy (%; GPT-4o judge; the model answers from the v | 100 |
Interactive version: theaggregate.ai/model?slug=qwen-3-5-35b-a3b · How It Works · Data refreshed daily, snapshot 2026-10-20.