Alibaba Cloud / Qwen Team model product
5-4B is a 4 billion parameter vision-language model using Gated DeltaNet hybrid architecture with a 3:1 ratio of linear attention to full softmax attention.
Updated Aug 12, 2026. Default version: Qwen3.5-4B
Technical details for the model's default version.
This profile uses the model's current scored version. Arena ratings and prices are shown separately.
Benchmark scores for Qwen3.5-4B.
| DeepPlanning | 17.6% | 09 | 9 | 0.0% | C | |
| MAXIFE | 78.0% | 09 | 11 | 20.0% | C | |
| NOVA-63 | 54.3% | 09 | 11 | 20.0% | C | |
| VITA-Bench | 22.0% | 10 | 10 | 0.0% | C | |
| BFCL-V4 | 50.3% | 11 | 13 | 16.7% | C | |
| Global PIQA | 78.9% | 11 | 13 | 16.7% | C | |
| t2-bench | 79.9% | 12 | 23 | 50.0% | C | |
| C-Eval | 85.1% | 13 | 18 | 29.4% | C | |
| PolyMATH | 51.1% | 13 | 23 | 45.5% | C | |
| WMT24++ | 66.6% | 13 | 23 | 45.5% | C | |
| AA-LCR | 57.0% | 14 | 16 | 13.3% | C | |
| HMMT25 | 76.8% | 14 | 25 | 45.8% | C | |
| LongBench v2 | 50.0% | 14 | 17 | 18.8% | C | |
| Multi-Challenge | 49.0% | 17 | 29 | 42.9% | C | |
| IFEval | 89.8% | 18 | 65 | 73.4% | C | |
| MMLU-ProX | 71.5% | 20 | 32 | 38.7% | C | |
| Include | 71.0% | 21 | 31 | 33.3% | C | |
| IFBench | 59.2% | 25 | 29 | 14.3% | C | |
| SuperGPQA | 52.9% | 26 | 34 | 24.2% | C | |
| HMMT 2025 | 74.0% | 27 | 33 | 18.8% | C | |
| MMLU-Redux | 88.8% | 29 | 48 | 40.4% | C | |
| LiveCodeBench v6 | 55.8% | 40 | 53 | 25.0% | C | |
| MMMLU | 76.1% | 42 | 49 | 14.6% | C | |
| MMLU-Pro | 79.1% | 56 | 129 | 57.0% | C | |
| GPQA | 76.2% | 100 | 234 | 57.5% | C |
Preference and agent-evaluation results for the default version.
The default version does not have a matching Arena result yet.
Provider-specific output speed and catalog latency for Qwen3.5-4B. Runtime does not affect the capability score.
No provider-specific speed or latency record is linked to the default version yet.
Output Speed is generated output tokens received per second. Catalog latency is reported separately from observed provider TTFT.
Official vendor API pricing appears first, followed by individual provider offers.
| QVAC | qwen3.5-4b | global | N/A | N/A | 32.8K | |
| EmpirioLabs AI | qwen3-5-4b | global | $0.04 | $0.07 | 262.1K |
Official prices use only the vendor's configured official Provider and positive standard USD PAYG rates. Third-party offers remain explicitly labeled.
Available versions of this model. The score column identifies the version used in the overall ranking.
| Qwen3.5-4B | 32.4 | 4B | N/A | N/A | No | Apache 2.0 |
Key information about Qwen3.5 4B and its available data.
5-4B is a 4 billion parameter vision-language model using Gated DeltaNet hybrid architecture with a 3:1 ratio of linear attention to full softmax attention. It supports 262K native context length and delivers strong performance for its size across knowledge, reasoning, coding, and multilingual tasks.
Data as of 2026-08-11.
Open a comparison with the three ranked models immediately above and below this model.
Recommendations prioritize the same model type and family, then the closest LLMBoard score.
Common questions about Qwen3.5 4B.
Qwen3.5 4B's default version was released on Mar 2, 2026.
No official standard PAYG price is currently available for Qwen3.5 4B. The lowest tracked third-party offer starts at $0.04 input and $0.07 output via EmpirioLabs AI.
Qwen3.5 4B was created by Alibaba Cloud / Qwen Team.
A context window is not available for the default version.
No. The default version is not marked as having publicly available weights.
2 provider offerings are linked to the default version.
Nearby ranked alternatives include DiffusionGemma 26B A4B, Mistral Small 4, LongCat Flash Lite.