Qwen 3 4B 2507 (Thinking): benchmark results
Qwen 3 4B 2507 evaluated with thinking enabled. Provider: Alibaba. Released 2025-08-06. Access: Open.
Aggregate Rating 1510 ± 9, rank #1037 of 2168 rated models, from 498 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| EuroEval Slovak NLU - UNER SK | 73.36 | Micro F1 excluding MISC (%) | 100 |
| FinMCP-Bench - Multi-Turn | 47.65 | Tool F1 (%): the harmonic mean of tool precision and tool re | 100 |
| FinMCP-Bench - Single-Tool | 68.55 | Tool F1 (%): the harmonic mean of tool precision and tool re | 100 |
| htkgh-polecat (Entity-Filtered History) | 40.6 | Relation prediction accuracy (%; choose the relation of a he | 100 |
| EuroEval German NLU - GermEval | 78.27 | Micro F1 excluding MISC (%) | 99.4 |
| EuroEval Dutch NLU - CoNLL NL | 80.14 | Micro F1 excluding MISC (%) | 98.8 |
| EuroEval Italian NLU - MultiNERD IT | 87.46 | Micro F1 excluding MISC (%) | 98.8 |
| EuroEval Spanish NLU | 60.1 | NLU Average Score (%) | 97.3 |
| EuroEval Latvian NLU - Fullstack NER LV | 73.2 | Micro F1 excluding MISC (%) | 97 |
| EuroEval Ukrainian NLU - NER UK | 79.05 | Micro F1 excluding MISC (%) | 97 |
| MERA Code - ruHumanEval | 78.11 | pass@1 (%) | 96.9 |
| EuroEval Norwegian NLU - NorNE NB | 86.13 | Micro F1 excluding MISC (%) | 96.7 |
Interactive version: theaggregate.ai/model?slug=qwen-3-4b-2507-thinking · How It Works · Data refreshed daily, snapshot 2026-10-20.