Alibaba Cloud / Qwen Team model product
Qwen3-VL-235B-A22B-Thinking is the most powerful vision-language model in the Qwen series, featuring 236B parameters with MoE architecture for reasoning-enhanced multimodal understanding. Key capabilities include: Visual Agent (operates PC/mobile GUIs, recognizes elements, invokes tools), Visual Coding (generates Draw.io/HTML/CSS/JS from images/videos), Advanced Spatial Perception (2D grounding and 3D grounding for spatial reasoning and embodied AI), Long Context & Video Understanding (native 256K context expandable to 1M, handles hours-long video with second-level indexing), Enhanced Multimodal Reasoning (excels in STEM/Math with causal analysis), Upgraded Visual Recognition (celebrities, anime, products, landmarks, flora/fauna), and Expanded OCR (32 languages, robust in low light/blur/tilt). Architecture innovations include Interleaved-MRoPE for positional embeddings, DeepStack for multi-level ViT feature fusion, and Text-Timestamp Alignment for precise video temporal modeling.
Updated Aug 10, 2026. Default version: Qwen3 VL 235B A22B Thinking
Structured fields from the published default version.
This profile uses the latest version under this unique model that has a calculated LLMBoard score. Arena and price are excluded.
Published benchmark records for the scored version Qwen3 VL 235B A22B Thinking.
| ARKitScenes | 0.5 | 1 | 1 | 100.0% | C | |
| LiveBench 20241125 | 0.8 | 1 | 14 | 100.0% | C | |
| MathVerse-Mini | 0.8 | 1 | 1 | 100.0% | C | |
| MIABench | 0.9 | 1 | 1 | 100.0% | C | |
| MMMUval | 0.8 | 1 | 4 | 100.0% | C | |
| Objectron | 0.7 | 1 | 1 | 100.0% | C | |
| OCRBench-V2 (zh) | 0.6 | 1 | 11 | 100.0% | C | |
| OSWorld-G | 0.7 | 1 | 1 | 100.0% | C | |
| RoboSpatialHome | 0.7 | 1 | 1 | 100.0% | C | |
| SIFO | 0.8 | 1 | 1 | 100.0% | C | |
| SIFO-Multiturn | 0.7 | 1 | 1 | 100.0% | C | |
| ZebraLogic | 1.0 | 1 | 8 | 100.0% | C | |
| CharadesSTA | 0.6 | 2 | 12 | 90.9% | C | |
| Design2Code | 0.9 | 2 | 2 | 0.0% | C | |
| InfoVQAtest | 0.9 | 2 | 12 | 90.9% | C | |
| MuirBench | 0.8 | 2 | 11 | 90.0% | C | |
| RefSpatialBench | 0.7 | 2 | 6 | 80.0% | C | |
| EmbSpatialBench | 0.8 | 3 | 8 | 71.4% | C | |
| Multi-IF | 0.8 | 3 | 20 | 89.5% | C | |
| RefCOCO-avg | 0.9 | 3 | 7 | 66.7% | C | |
| SUNRGBD | 0.3 | 3 | 4 | 33.3% | C | |
| VisuLogic | 0.3 | 3 | 3 | 0.0% | C | |
| WritingBench | 0.9 | 3 | 15 | 85.7% | C | |
| DocVQAtest | 1.0 | 4 | 11 | 70.0% | C | |
| Hypersim | 0.1 | 4 | 4 | 0.0% | C | |
| OCRBench-V2 (en) | 0.7 | 4 | 12 | 72.7% | C | |
| ScreenSpot | 1.0 | 4 | 16 | 80.0% | C | |
| CC-OCR | 0.8 | 5 | 18 | 76.5% | C | |
| Creative Writing v3 | 0.9 | 5 | 13 | 66.7% | C | |
| MMLongBench-Doc | 0.6 | 5 | 5 | 0.0% | C | |
| MMLU | 0.9 | 5 | 100 | 96.0% | C | |
| ZEROBench-Sub | 0.3 | 5 | 5 | 0.0% | C | |
| CountBench | 0.9 | 6 | 6 | 0.0% | C | |
| Hallusion Bench | 0.7 | 6 | 16 | 66.7% | C | |
| MM-MT-Bench | 8.5 | 6 | 17 | 68.8% | C | |
| VideoMME w/o sub. | 0.8 | 6 | 10 | 44.4% | C | |
| BFCL-v3 | 0.7 | 7 | 19 | 66.7% | C | |
| MMBench-V1.1 | 0.9 | 7 | 18 | 64.7% | C | |
| MMStar | 0.8 | 7 | 22 | 71.4% | C | |
| MathVista-Mini | 0.9 | 8 | 23 | 68.2% | C | |
| MMLU-Redux | 0.9 | 8 | 48 | 85.1% | C | |
| BLINK | 0.7 | 9 | 13 | 33.3% | C | |
| MLVU | 0.8 | 9 | 10 | 11.1% | C | |
| SimpleVQA | 0.6 | 9 | 13 | 33.3% | C | |
| ZEROBench | 0.0 | 9 | 9 | 0.0% | C | |
| Include | 0.8 | 10 | 31 | 70.0% | C | |
| MMLU-ProX | 0.8 | 10 | 32 | 71.0% | C | |
| ODinW | 0.4 | 10 | 16 | 40.0% | C | |
| OSWorld | 0.4 | 11 | 20 | 47.4% | C | |
| RealWorldQA | 0.8 | 11 | 26 | 60.0% | C | |
| LVBench | 0.6 | 12 | 24 | 52.2% | C | |
| ERQA | 0.5 | 13 | 23 | 45.5% | C | |
| HMMT25 | 0.8 | 13 | 25 | 50.0% | C | |
| OCRBench | 0.9 | 13 | 22 | 42.9% | C | |
| ScreenSpot Pro | 0.6 | 13 | 24 | 47.8% | C | |
| SuperGPQA | 0.6 | 13 | 34 | 63.6% | C | |
| AI2D | 0.9 | 14 | 32 | 58.1% | C | |
| MathVision | 0.7 | 14 | 32 | 58.1% | C | |
| VideoMMMU | 0.8 | 17 | 26 | 36.0% | C | |
| SimpleQA | 0.4 | 18 | 46 | 62.2% | C | |
| IFEval | 0.9 | 27 | 65 | 59.4% | C | |
| MMLU-Pro | 0.8 | 28 | 129 | 78.9% | C | |
| LiveCodeBench v6 | 0.7 | 29 | 53 | 46.1% | C | |
| CharXiv-R | 0.7 | 32 | 47 | 32.6% | C | |
| MMMU-Pro | 0.7 | 36 | 65 | 45.3% | C | |
| AIME 2025 | 0.9 | 49 | 114 | 57.5% | C | |
| Humanity's Last Exam | 0.1 | 75 | 92 | 18.7% | C |
Preference and agent-evaluation signals from published Arena datasets.
| vision | creative writing | 13 | 1229.9 | 219 | N/A | |
| vision style control | creative writing | 14 | 1217.6 | 219 | N/A | |
| vision | homework | 62 | 1235.0 | 306 | N/A | |
| vision | english | 68 | 1219.4 | 1,096 | N/A | |
| vision | diagram | 69 | 1213.4 | 476 | N/A | |
| vision | ocr | 69 | 1212.5 | 1,347 | N/A | |
| vision style control | homework | 69 | 1227.3 | 306 | N/A | |
| vision | overall | 70 | 1208.2 | 2,347 | N/A | |
| text | industry mathematical | 73 | 1437.3 | 377 | N/A | |
| vision style control | english | 73 | 1197.2 | 1,096 | N/A | |
| vision style control | diagram | 75 | 1209.6 | 476 | N/A | |
| vision style control | ocr | 75 | 1201.2 | 1,347 | N/A | |
| vision style control | overall | 77 | 1189.8 | 2,347 | N/A | |
| text | german | 79 | 1422.1 | 170 | N/A | |
| text style control | german | 92 | 1420.3 | 170 | N/A | |
| text | korean | 95 | 1363.0 | 211 | N/A | |
| text style control | industry mathematical | 98 | 1431.7 | 377 | N/A | |
| text style control | korean | 102 | 1362.9 | 211 | N/A | |
| text | chinese | 109 | 1451.4 | 296 | N/A | |
| text | expert | 111 | 1418.8 | 379 | N/A | |
| text | math | 112 | 1414.0 | 424 | N/A | |
| text | coding | 115 | 1427.5 | 1,627 | N/A | |
| text | industry medicine and healthcare | 116 | 1416.3 | 433 | N/A | |
| text style control | expert | 119 | 1430.3 | 379 | N/A | |
| text | english | 120 | 1417.7 | 3,882 | N/A | |
| text | industry entertainment and sports and media | 120 | 1370.6 | 1,440 | N/A | |
| text | industry software and it services | 120 | 1426.9 | 2,834 | N/A | |
| text | polish | 124 | 1378.8 | 389 | N/A | |
| text | hard prompts english | 125 | 1415.5 | 2,094 | N/A | |
| text | industry legal and government | 125 | 1408.1 | 513 | N/A | |
| text | spanish | 126 | 1384.9 | 330 | N/A | |
| text style control | chinese | 126 | 1438.7 | 296 | N/A | |
| text style control | coding | 126 | 1454.6 | 1,627 | N/A | |
| text | hard prompts | 128 | 1407.8 | 4,039 | N/A | |
| text | industry business and management and financial operations | 129 | 1391.6 | 1,580 | N/A | |
| text style control | math | 129 | 1404.0 | 424 | N/A | |
| text | overall | 130 | 1400.9 | 7,934 | N/A | |
| text | exclude ties | 130 | 1379.8 | 5,579 | N/A | |
| text style control | industry software and it services | 133 | 1436.9 | 2,834 | N/A | |
| text style control | polish | 133 | 1376.9 | 389 | N/A | |
| text | industry life and physical and social science | 134 | 1411.8 | 1,247 | N/A | |
| text | longer query | 134 | 1393.6 | 1,725 | N/A | |
| text style control | spanish | 134 | 1384.4 | 330 | N/A | |
| text style control | hard prompts | 137 | 1418.7 | 4,039 | N/A | |
| text style control | non english | 137 | 1381.8 | 4,052 | N/A | |
| text | non english | 138 | 1378.8 | 4,052 | N/A | |
| text style control | industry legal and government | 138 | 1404.3 | 513 | N/A | |
| text style control | longer query | 138 | 1406.7 | 1,725 | N/A | |
| text | instruction following | 140 | 1374.4 | 2,136 | N/A | |
| text | multi turn | 140 | 1389.7 | 1,288 | N/A | |
| text style control | industry entertainment and sports and media | 140 | 1361.6 | 1,440 | N/A | |
| text style control | overall | 142 | 1395.4 | 7,934 | N/A | |
| text | industry writing and literature and language | 143 | 1365.2 | 1,752 | N/A | |
| text style control | industry medicine and healthcare | 143 | 1415.6 | 433 | N/A | |
| text style control | industry business and management and financial operations | 144 | 1393.0 | 1,580 | N/A | |
| text style control | exclude ties | 146 | 1372.7 | 5,579 | N/A | |
| text style control | instruction following | 146 | 1383.1 | 2,136 | N/A | |
| text style control | russian | 148 | 1377.5 | 481 | N/A | |
| text | creative writing | 149 | 1344.4 | 1,026 | N/A | |
| text style control | industry life and physical and social science | 149 | 1406.4 | 1,247 | N/A | |
| text style control | english | 150 | 1403.3 | 3,882 | N/A | |
| text style control | hard prompts english | 150 | 1419.6 | 2,094 | N/A | |
| text | russian | 152 | 1360.4 | 481 | N/A | |
| text style control | industry writing and literature and language | 154 | 1365.2 | 1,752 | N/A | |
| text style control | multi turn | 158 | 1387.0 | 1,288 | N/A | |
| text style control | creative writing | 167 | 1338.5 | 1,026 | N/A |
Official vendor API PAYG pricing is summarized first. The table then lists individual provider offerings without treating their minimum as the official price.
| OpenRouter | qwen/qwen3-vl-235b-a22b-thinking | global | $0.40 | $4 | 131.1K | |
| NanoGPT | qwen3-vl-235b-a22b-thinking | global | $0.50 | $6 | 32.8K | |
| LLM Gateway | qwen3-vl-235b-a22b-thinking | global | $0.98 | $3.95 | 131.1K | |
| NovitaAI | qwen/qwen3-vl-235b-a22b-thinking | global | $0.98 | $3.95 | 131.1K |
Official prices use only the vendor's configured official Provider and positive standard USD PAYG rates. Third-party offers remain explicitly labeled.
All published versions linked to this unique model. The score columns identify the version used by the current overall ranking.
| Qwen3 VL 235B A22B Thinking | 58.5 | 236B | 262.1K | 262.1K | No | Apache 2.0 |
A concise description based on the published model registry.
Qwen3-VL-235B-A22B-Thinking is the most powerful vision-language model in the Qwen series, featuring 236B parameters with MoE architecture for reasoning-enhanced multimodal understanding. Key capabilities include: Visual Agent (operates PC/mobile GUIs, recognizes elements, invokes tools), Visual Coding (generates Draw.io/HTML/CSS/JS from images/videos), Advanced Spatial Perception (2D grounding and 3D grounding for spatial reasoning and embodied AI), Long Context & Video Understanding (native 256K context expandable to 1M, handles hours-long video with second-level indexing), Enhanced Multimodal Reasoning (excels in STEM/Math with causal analysis), Upgraded Visual Recognition (celebrities, anime, products, landmarks, flora/fauna), and Expanded OCR (32 languages, robust in low light/blur/tilt). Architecture innovations include Interleaved-MRoPE for positional embeddings, DeepStack for multi-level ViT feature fusion, and Text-Timestamp Alignment for precise video temporal modeling.
Use the benchmark, Arena and pricing sections above as separate evidence. A missing field means the current data snapshot does not support that claim.
Data snapshot: 2026-08-07. Editorial model content is not available in the backend.
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 published LLMBoard score.
Common questions about Qwen3 VL 235B A22B Thinking.
Qwen3 VL 235B A22B Thinking's default version was released on Sep 22, 2025.
No official standard PAYG price is currently available for Qwen3 VL 235B A22B Thinking. The lowest tracked third-party offer starts at $0.40 input and $4 output via OpenRouter.
Qwen3 VL 235B A22B Thinking is published under Alibaba Cloud / Qwen Team in the model registry.
The default version has a 262.1K token context window.
No. The default version is not marked as having publicly available weights.
4 published provider offerings are linked to the default version.
Nearby ranked alternatives include Nemotron 3 Super, Claude Opus 3, GPT-4.5.