Qwen3 VL 32B 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 | 89.5% | self-reported llm-stats | link → |
| AIME 2025 | 66.2% | self-reported llm-stats | link → |
| Arena-Hard v2 | 64.7% | self-reported llm-stats | link → |
| BFCL-v3 | 70.2% | self-reported llm-stats | link → |
| BLINK | 67.3% | self-reported llm-stats | link → |
| CC-OCR | 80.3% | self-reported llm-stats | link → |
| CharadesSTA | 61.2% | self-reported llm-stats | link → |
| CharXiv-D | 90.5% | self-reported llm-stats | link → |
| CharXiv-R | 62.8% | self-reported llm-stats | link → |
| Creative Writing v3 | 0.856 | self-reported llm-stats | link → |
| DocVQAtest | 96.9% | self-reported llm-stats | link → |
| ERQA | 48.8% | self-reported llm-stats | link → |
| GPQA | 68.9% | self-reported llm-stats | link → |
| Hallusion Bench | 63.8% | self-reported llm-stats | link → |
| IFEval | 84.7% | self-reported llm-stats | link → |
| Include | 74.0% | self-reported llm-stats | link → |
| InfoVQAtest | 87.0% | self-reported llm-stats | link → |
| LiveBench 20241125 | 72.2% | self-reported llm-stats | link → |
| LiveCodeBench v6 | 43.8% | self-reported llm-stats | link → |
| LVBench | 63.8% | self-reported llm-stats | link → |
| MathVision | 63.4% | self-reported llm-stats | link → |
| MathVista-Mini | 83.8% | self-reported llm-stats | link → |
| MLVU-M | 82.1% | self-reported llm-stats | link → |
| MM-MT-Bench | 8.4 | self-reported llm-stats | link → |
| MMLU | 86.4% | self-reported llm-stats | link → |
| MMLU-Pro | 78.6% | self-reported llm-stats | link → |
| MMLU-ProX | 73.4% | self-reported llm-stats | link → |
| MMLU-Redux | 89.8% | self-reported llm-stats | link → |
| MMMU (val) | 76.0% | self-reported llm-stats | link → |
| MMMU-Pro | 65.3% | self-reported llm-stats | link → |
| MMStar | 77.7% | self-reported llm-stats | link → |
| MuirBench | 72.8% | self-reported llm-stats | link → |
| Multi-IF | 72.0% | self-reported llm-stats | link → |
| MVBench | 72.8% | self-reported llm-stats | link → |
| OCRBench | 89.5% | self-reported llm-stats | link → |
| OCRBench-V2 (en) | 67.4% | self-reported llm-stats | link → |
| OCRBench-V2 (zh) | 59.2% | self-reported llm-stats | link → |
| ODinW | 46.6% | self-reported llm-stats | link → |
| OSWorld | 32.6% | self-reported llm-stats | link → |
| PolyMATH | 40.5% | self-reported llm-stats | link → |
| RealWorldQA | 79.0% | self-reported llm-stats | link → |
| ScreenSpot | 95.8% | self-reported llm-stats | link → |
| ScreenSpot Pro | 57.9% | self-reported llm-stats | link → |
| SuperGPQA | 54.6% | self-reported llm-stats | link → |
| WritingBench | 82.9% | self-reported llm-stats | link → |