Qwen3 VL 32B Thinking
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 | 88.9% | self-reported llm-stats | link → |
| AIME 2025 | 83.7% | self-reported llm-stats | link → |
| AndroidWorld_SR | 63.7% | self-reported llm-stats | link → |
| Arena-Hard v2 | 60.5% | self-reported llm-stats | link → |
| BFCL-v3 | 71.7% | self-reported llm-stats | link → |
| BLINK | 68.5% | self-reported llm-stats | link → |
| CharadesSTA | 62.8% | self-reported llm-stats | link → |
| CharXiv-D | 90.2% | self-reported llm-stats | link → |
| CharXiv-R | 65.2% | self-reported llm-stats | link → |
| Creative Writing v3 | 0.833 | self-reported llm-stats | link → |
| DocVQAtest | 96.1% | self-reported llm-stats | link → |
| ERQA | 52.3% | self-reported llm-stats | link → |
| GPQA | 73.1% | self-reported llm-stats | link → |
| Hallusion Bench | 67.4% | self-reported llm-stats | link → |
| IFEval | 87.8% | self-reported llm-stats | link → |
| Include | 76.3% | self-reported llm-stats | link → |
| InfoVQAtest | 89.2% | self-reported llm-stats | link → |
| LiveBench 20241125 | 74.7% | self-reported llm-stats | link → |
| LiveCodeBench v6 | 65.6% | self-reported llm-stats | link → |
| LVBench | 62.6% | self-reported llm-stats | link → |
| MathVision | 70.2% | self-reported llm-stats | link → |
| MathVista-Mini | 85.9% | self-reported llm-stats | link → |
| MM-MT-Bench | 8.3 | self-reported llm-stats | link → |
| MMBench-V1.1 | 90.8% | self-reported llm-stats | link → |
| MMLU | 88.7% | self-reported llm-stats | link → |
| MMLU-Pro | 82.1% | self-reported llm-stats | link → |
| MMLU-ProX | 77.2% | self-reported llm-stats | link → |
| MMLU-Redux | 91.9% | self-reported llm-stats | link → |
| MMMU (val) | 78.1% | self-reported llm-stats | link → |
| MMMU-Pro | 68.1% | self-reported llm-stats | link → |
| MMStar | 79.4% | self-reported llm-stats | link → |
| MuirBench | 80.3% | self-reported llm-stats | link → |
| Multi-IF | 78.0% | self-reported llm-stats | link → |
| MVBench | 73.2% | self-reported llm-stats | link → |
| OCRBench | 85.5% | self-reported llm-stats | link → |
| OCRBench-V2 (en) | 68.4% | self-reported llm-stats | link → |
| OCRBench-V2 (zh) | 62.1% | self-reported llm-stats | link → |
| OSWorld | 41.0% | self-reported llm-stats | link → |
| PolyMATH | 52.0% | self-reported llm-stats | link → |
| RealWorldQA | 78.4% | self-reported llm-stats | link → |
| ScreenSpot | 95.7% | self-reported llm-stats | link → |
| ScreenSpot Pro | 57.1% | self-reported llm-stats | link → |
| SimpleQA | 55.4% | self-reported llm-stats | link → |
| SuperGPQA | 59.0% | self-reported llm-stats | link → |
| VideoMME w/o sub. | 77.3% | self-reported llm-stats | link → |
| VideoMMMU | 79.0% | self-reported llm-stats | link → |
| WritingBench | 86.2% | self-reported llm-stats | link → |