Qwen 3 4B (Reasoning): benchmark results
Qwen 3 4B evaluated with reasoning enabled. Provider: Alibaba. Released 2025-04-29. Access: Open.
Aggregate Rating 1482 ± 11, rank #1150 of 2170 rated models, from 141 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| PersonalHomeBench - Multiple Choice (Sole Reasoning) | 90.3 | Accuracy (%) on 1,000 multiple-choice household questions (e | 88.9 |
| WildToolBench - Single-Tool Tasks | 60.16 | Accuracy (%) on tasks solvable with one tool call, on WildTo | 85.7 |
| Avalon-ToM-Bench - Intention Attribution | 86.27 | Accuracy (%) | 76.9 |
| WildToolBench - Task Accuracy | 53.91 | Task accuracy (%) over all 1,024 tasks, on WildToolBench (25 | 76.8 |
| Medmarks - M-ARC | 57.67 | Score (%) | 75 |
| WildToolBench - Clarification Tasks | 38.67 | Accuracy (%) on tasks where the correct response is to ask t | 71.4 |
| BRIDGE Medical Leaderboard - CoT | 36.98 | Average Performance (%) | 69.4 |
| Swallow - Japanese MT-Bench - Extraction | 73.4 | Judge Score (normalized, %) | 69.4 |
| Avalon-ToM-Bench - Strategic Signaling | 76.87 | Accuracy (%) | 69.2 |
| Swallow - English MT-Bench - Reasoning | 84.1 | Judge Score (normalized, %) | 67.2 |
| Medmarks - MedHallu Easy | 67.89 | Score (%) | 65.9 |
| Medmarks - MedHallu Hard | 47.93 | Score (%) | 65.9 |
Interactive version: theaggregate.ai/model?slug=qwen-3-4b-reasoning · How It Works · Data refreshed daily, snapshot 2026-11-05.