Qwen 1.5 4B Chat: benchmark results
Alibaba's Qwen1.5-4B-Chat, the aligned chat version of Qwen1.5-4B, post-trained with supervised fine-tuning and direct preference optimization, supporting a 32K context window. Provider: Alibaba. Released 2024-02-04. Access: Open.
Aggregate Rating 1307 ± 14, rank #1467 of 1656 rated models, from 166 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| SALAD-Bench | 59.43 | Average Safety Score (%) | 78.8 |
| SALAD-Bench Attack | 23.34 | Safety Score (%) | 75.8 |
| EuroEval Portuguese NLU - MultiWikiQA PT | 73.03 | Character F1 (%) | 74.4 |
| SALAD-Bench Base | 95.51 | Safety Score (%) | 72.7 |
| OpenCompass Agent - CIBench - Chinese (CompassBench 2402) | 41.6 | Score (%) | 64.7 |
| OpenCompass LLM - Agent - Chinese (CompassBench 2402) | 52.4 | Score (%) | 61.8 |
| OpenCompass LLM - Agent (CompassBench 2402) | 52.6 | Score (%) | 55.9 |
| RewardBench Chat Hard | 62.72 | Accuracy (%) | 53.7 |
| Open Chinese LLM - WinoGrande | 63.14 | Accuracy (%) | 53.4 |
| OpenCompass LLM - Agent - English (CompassBench 2402) | 52.7 | Score (%) | 50 |
| MERA - RWSD | 55.77 | Accuracy (%) | 48.3 |
| OpenCompass Agent - CIBench - English (CompassBench 2402) | 43.6 | Score (%) | 47.1 |
Interactive version: theaggregate.ai/model?slug=qwen-1-5-4b-chat · How It Works · Data refreshed daily, snapshot 2026-10-20.