Qwen 3 4B: benchmark results
Can switch between direct answers and explicit reasoning within the same conversation, without loading another checkpoint. Training covers multilingual dialogue, code and tool use. Provider: Alibaba. Released 2025-04-29. Access: Open.
Aggregate Rating 1485 ± 5, rank #822 of 1635 rated models, from 1345 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| LingxiDiagBench (Dynamic, Clinical, APA-Guided + MRD-RAG) - 12-Class Accuracy | 46.5 | Accuracy (%, 0-100) on twelve-category ICD-10 diagnosis (F20 | 100 |
| LingxiDiagBench (Static, Clinical) - 2-Class Accuracy | 88.7 | Accuracy (times 100, 0-100) on binary depression-versus-anxi | 100 |
| LingxiDiagBench (Static, Clinical) - Next-Question BERTScore | 77.6 | BERTScore (%, 0-100) of the predicted next doctor question a | 100 |
| ProtStructQA | 70.33 | Standard (self-reported) | 100 |
| Syslog Severity Classification - Few-Shot | 56.01 | Accuracy on 9,355 held-out journalctl logs, exact syslog sev | 100 |
| Syslog Severity Classification - RAG | 95.64 | Accuracy on 9,355 held-out journalctl logs, exact syslog sev | 100 |
| TRACK - Math Reasoning (MeLLo) | 57 | Holistic pass (%; PRM800K-derived problems decomposed into p | 100 |
| EuroEval Ukrainian Summarization - LR SUM UK | 31.56 | chrF3++ (%) | 97.2 |
| Open Portuguese LLM - BLUEX | 78.03 | Accuracy (%) | 97.1 |
| Open Portuguese LLM - ENEM | 84.11 | Accuracy (%) | 97 |
| MERA - ruHumanEval | 76.16 | pass@1 (%) | 95.7 |
| EuroEval Albanian Summarization - LR SUM SQ | 34.22 | chrF3++ (%) | 95.4 |
Interactive version: theaggregate.ai/model?slug=qwen-3-4b · How It Works · Data refreshed daily, snapshot 2026-10-15.