Qwen 3.6 35B A3B: benchmark results
Alibaba's efficiency-focused multimodal Qwen 3.6 35B A3B MoE, with 3B active parameters. Provider: Alibaba. Released 2026-04-16. Access: Open.
Aggregate Rating 1690 ± 5, rank #217 of 1640 rated models, from 1106 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| AfriEconQA (Oracle Evidence) | 48.9 | Token F1 (%; 862 held-out test questions, answered from the | 100 |
| AsyncPlan Robo Challenge (PDDL2.1 Formalizer) | 67.9 | Plan accuracy (%; share of the 140 Robo Challenge kitchen ta | 100 |
| AsyncPlan Robo Challenge (Planner) | 59.3 | Plan accuracy (%; share of the 140 Robo Challenge kitchen ta | 100 |
| AsyncPlan-XXL (Planner) | 70 | Plan accuracy (%; mean over the 12 AsyncPlan-XXL graph sizes | 100 |
| FormStruct-Bench - Path Accuracy | 17.91 | TSR-path (%; share of ground-truth leaf field-path and value | 100 |
| MERA Multi - ruMathVQA | 40.1 | Score (%) | 100 |
| MMHBench - First-Person | 53.7 | Accuracy (%; the 1,579 first-person perspective-taking multi | 100 |
| Nejumi 4 - ALT - Bias | 100 | Score (%) | 100 |
| SEA-HELM (Indonesian) - Sentiment Analysis | 81.84 | Normalized Score | 100 |
| SEA-HELM (Tamil) - Linguistic Diagnostics | 85.18 | Normalized Score | 100 |
| SEA-HELM (Tamil) - Pragmatics | 80.83 | Normalized Score | 100 |
| SEA-HELM (Thai) - Natural Language Inference | 61.28 | Normalized Score | 100 |
Interactive version: theaggregate.ai/model?slug=qwen-3-6-35b-a3b · How It Works · Data refreshed daily, snapshot 2026-10-13.