Qwen 3.5 4B: benchmark results
Alibaba's open 4B from the Qwen3.5 small series (March 2026): natively multimodal with 262K context, built for on-device use. Provider: Alibaba. Released 2026-03-02. Access: Open.
Aggregate Rating 1590 ± 5, rank #475 of 1656 rated models, from 1485 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Bangladesh Legal QA (No Law) | 76.7 | Accuracy (%; no legal context supplied, parametric knowledge | 100 |
| Bangladesh Legal QA (Supplied Law) | 94 | Accuracy (%; the governing statutory section is supplied wit | 100 |
| KIDBench (No Cue) - Physical Health and Safety | 3.02 | Category score (1-5): mean of the five child-safety dimensio | 100 |
| Kurakurai French LLM - AIME 2024 | 58.33 | Accuracy (%) | 100 |
| Kurakurai French LLM - BFCL v2 | 73.61 | Accuracy (%) | 100 |
| Kurakurai French LLM - GPQA Diamond | 64.3 | Accuracy (%) | 100 |
| Kurakurai French LLM - Global MMLU | 76.5 | Accuracy (%) | 100 |
| Kurakurai French LLM - HumanEval+ | 79.75 | pass@1 (%) | 100 |
| Kurakurai French LLM - MMLU-Pro | 71.4 | Accuracy (%) | 100 |
| MCJudgeBench - Response-Variant Inconsistency | 3.59 | Procedural inconsistency rate under candidate-response varia | 100 |
| Kurakurai French LLM - AIME 2025 | 37 | Accuracy (%) | 97.6 |
| Kurakurai French LLM - MATH-500 | 89.26 | Accuracy (%) | 97.6 |
Interactive version: theaggregate.ai/model?slug=qwen-3-5-4b · How It Works · Data refreshed daily, snapshot 2026-10-20.