Qwen3 VL 4B Instruct
Qwen3-VL is a large multimodal model that unifies vision, language, and reasoning to achieve human-level perception and cognition across text, images, and video. Built on a 235B-parameter architecture, it integrates early joint training of visual and textual modalities for strong language grounding. The model supports up to a 1 million-token context window and excels at visual understanding, spatial reasoning, long video comprehension, and tool-based interaction. It can generate code from images, perform precise 2D/3D object grounding, and operate digital interfaces like a visual agent. The “Instruct” version rivals Gemini 2.5 Pro in perception benchmarks, while the “Thinking” version leads in multimodal reasoning and STEM tasks. With multilingual OCR, creative writing, and fine-grained scene interpretation, Qwen3-VL establishes a new open-source frontier for integrated vision-language intelligence.
Benchmark results
| Benchmark | Score | Tags | Source |
|---|---|---|---|
| AI2D | 84.1% | self-reported llm-stats | link → |
| AIME 2025 | 46.6% | self-reported llm-stats | link → |
| BFCL-v3 | 63.3% | self-reported llm-stats | link → |
| BLINK | 65.8% | self-reported llm-stats | link → |
| CC-OCR | 76.2% | self-reported llm-stats | link → |
| CharadesSTA | 55.5% | self-reported llm-stats | link → |
| CharXiv-D | 76.2% | self-reported llm-stats | link → |
| CharXiv-R | 39.7% | self-reported llm-stats | link → |
| DocVQAtest | 95.3% | self-reported llm-stats | link → |
| ERQA | 41.3% | self-reported llm-stats | link → |
| Hallusion Bench | 57.6% | self-reported llm-stats | link → |
| HMMT25 | 30.7% | self-reported llm-stats | link → |
| IFEval | 82.3% | self-reported llm-stats | link → |
| Include | 61.4% | self-reported llm-stats | link → |
| InfoVQAtest | 80.3% | self-reported llm-stats | link → |
| LiveBench 20241125 | 60.9% | self-reported llm-stats | link → |
| LiveCodeBench v6 | 37.9% | self-reported llm-stats | link → |
| LVBench | 56.2% | self-reported llm-stats | link → |
| MathVision | 51.6% | self-reported llm-stats | link → |
| MathVista-Mini | 73.7% | self-reported llm-stats | link → |
| MLVU-M | 75.3% | self-reported llm-stats | link → |
| MM-MT-Bench | 7.5 | self-reported llm-stats | link → |
| MMBench-V1.1 | 85.1% | self-reported llm-stats | link → |
| MMLU | 77.2% | self-reported llm-stats | link → |
| MMLU-Pro | 67.1% | self-reported llm-stats | link → |
| MMLU-ProX | 59.4% | self-reported llm-stats | link → |
| MMLU-Redux | 81.5% | self-reported llm-stats | link → |
| MMMU (val) | 67.4% | self-reported llm-stats | link → |
| MMMU-Pro | 53.2% | self-reported llm-stats | link → |
| MMStar | 69.8% | self-reported llm-stats | link → |
| MuirBench | 63.8% | self-reported llm-stats | link → |
| MVBench | 68.9% | self-reported llm-stats | link → |
| OCRBench | 88.1% | self-reported llm-stats | link → |
| OCRBench-V2 (en) | 63.7% | self-reported llm-stats | link → |
| OCRBench-V2 (zh) | 57.6% | self-reported llm-stats | link → |
| ODinW | 48.2% | self-reported llm-stats | link → |
| OSWorld | 26.2% | self-reported llm-stats | link → |
| PolyMATH | 28.8% | self-reported llm-stats | link → |
| RealWorldQA | 70.9% | self-reported llm-stats | link → |
| ScreenSpot | 94.0% | self-reported llm-stats | link → |
| ScreenSpot Pro | 59.5% | self-reported llm-stats | link → |
| SimpleQA | 48.0% | self-reported llm-stats | link → |
| SuperGPQA | 40.3% | self-reported llm-stats | link → |
| VideoMMMU | 56.2% | self-reported llm-stats | link → |
| WritingBench | 82.5% | self-reported llm-stats | link → |