Qwen3.5-4B-Base: benchmark results
Alibaba's Qwen3.5-4B-Base, the pretrained-only checkpoint in the Qwen3.5 series, a multimodal vision-language foundation model using a hybrid Gated Delta Network and sparse mixture-of-experts architecture. Provider: Alibaba. Released 2026-03-01. Access: Open.
Aggregate Rating 1494 ± 20, rank #801 of 1640 rated models, from 293 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| EuroEval Faroese NLU - FoQA | 82.59 | Character F1 (%) | 100 |
| EuroEval Finnish NLU - Tydiqa FI | 77.51 | Character F1 (%) | 99.5 |
| EuroEval Icelandic NLU - NQII | 65.37 | Character F1 (%) | 99.5 |
| EuroEval Italian NLU - SQuAD IT | 78.05 | Character F1 (%) | 99.3 |
| EuroEval Czech NLU - SQAD | 72.57 | Character F1 (%) | 98.7 |
| EuroEval Bosnian NLU - Multi Wiki QA BS | 76.27 | Character F1 (%) | 98.6 |
| EuroEval Bulgarian NLU - Cinexio | 61.35 | Matthews correlation coefficient (%) | 98.6 |
| EuroEval Dutch NLU - SQuAD NL | 84.17 | Character F1 (%) | 98.5 |
| EuroEval Polish NLU - PoQuAD | 65.76 | Character F1 (%) | 98.3 |
| EuroEval Estonian NLU - MultiWikiQA ET | 75.36 | Character F1 (%) | 98 |
| EuroEval German NLU - Germanquad | 68.99 | Character F1 (%) | 97.8 |
| EuroEval Swedish NLU - Multi Wiki QA SV | 80.75 | Character F1 (%) | 97.2 |
Interactive version: theaggregate.ai/model?slug=qwen3-5-4b-base · How It Works · Data refreshed daily, snapshot 2026-11-07.