Qwen 3.5 4B (Thinking): benchmark results
Provider: Alibaba. Released 2026-03-02. Access: Open.
Aggregate Rating 1547 ± 7, rank #891 of 2170 rated models, from 263 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| LFM3-VL-2B Release - DocVQA (val) | 97.1 | Published normalized score (0–100) | 100 |
| LFM3-VL-2B Release - HR-Bench (4K) | 88.1 | Published normalized score (0–100) | 100 |
| LFM3-VL-2B Release - HR-Bench (8K) | 82.1 | Published normalized score (0–100) | 100 |
| LFM3-VL-2B Release - MathVista (mini) | 84.8 | Published normalized score (0–100) | 100 |
| LFM3-VL-2B Release - MuirBench | 72.8 | Published normalized score (0–100) | 100 |
| LFM3-VL-2B Release - RealWorldQA | 81 | Published normalized score (0–100) | 100 |
| LFM3-VL-2B Release - RoboSpatial-Home | 67.4 | Published normalized score (0–100) | 100 |
| LFM3-VL-2B Release - V*Bench | 89.5 | Published normalized score (0–100) | 100 |
| LFM3-VL-2B Release - We-Math (testmini, strict) | 81 | Published normalized score (0–100) | 100 |
| LFM3-VL-2B Release - CV-Bench | 88.4 | Published normalized score (0–100) | 90 |
| LFM3-VL-2B Release - EmbSpatial-Bench | 82.4 | Published normalized score (0–100) | 90 |
| LFM3-VL-2B Release - HallusionBench | 66.2 | Published normalized score (0–100) | 90 |
Interactive version: theaggregate.ai/model?slug=qwen-3-5-4b-thinking · How It Works · Data refreshed daily, snapshot 2026-11-05.