Qwen3-VL-4B
Qwen3-VL-4B is an open-weights language model from Qwen, released October 15, 2025. It has 4.4B parameters (dense). The context window is 256K tokens.
We introduce Qwen3-VL, the most capable vision-language model in the Qwen series to date, achieving superior performance across a broad range of multimodal benchmarks. It natively supports interleaved contexts of up to 256K tokens, seamlessly integrating text, images, and video.
Check it against your own PC →
Can you run Qwen3-VL-4B locally?
The smallest common GPU it runs on is 6 GB of VRAM (with 32 GB of RAM): Q3_K_M, fully on GPU, roughly 50 tokens per second. Estimates assume 32 GB of DDR5 RAM, a 16K context and llama.cpp-style GGUF quantization. Enter your exact hardware in the live app.
| GPU memory | Verdict | Best quant | VRAM | RAM | tok/s |
|---|---|---|---|---|---|
| 6 GB VRAM | Runs on GPU | Q3_K_M (okay) | 5.2 GB | – | ~50 |
| 8 GB VRAM | Runs on GPU | Q6_K (≈ lossless) | 6.7 GB | – | ~43 |
| 12 GB VRAM | Runs on GPU | Q6_K (≈ lossless) | 6.7 GB | – | ~56 |
| 16 GB VRAM | Runs on GPU | Q6_K (≈ lossless) | 6.7 GB | – | ~76 |
| 24 GB VRAM | Runs on GPU | Q6_K (≈ lossless) | 6.7 GB | – | ~130 |
| 32 GB VRAM | Runs on GPU | Q6_K (≈ lossless) | 6.7 GB | – | ~230 |
Ready-made quantizations on Hugging Face: Qwen/Qwen3-VL-4B-Instruct-GGUF · unsloth/Qwen3-VL-4B-Instruct-GGUF · lmstudio-community/Qwen3-VL-4B-Instruct-GGUF · bartowski/Qwen_Qwen3-VL-4B-Instruct-GGUF · off-grid-ai/Video-Companions-GGUF · lmstudio-community/Qwen3-VL-4B-Instruct-MLX-4bit
Benchmark results
Epoch AI has not published results for Qwen3-VL-4B yet. New models are usually evaluated within days to weeks.
Links
Announcement · Hugging Face: Qwen/Qwen3-VL-4B-Instruct
More from Qwen
Qwen3.8 Omni Flash · Qwen3.8 Max (0902) · Qwen3.8 Flash · Qwen3.8 Flash Next · Qwen 3.8 27B · Qwen3.8-2.4T-A95B · Qwen 3.8 Max · Qwen3.7 Flash