[whatmodel.sh](../../README.md) / Qwen3-VL 30B-A3B Thinking # Qwen3-VL 30B-A3B Thinking Alibaba · released 2025-09 · 31.1B params (3B active) · apache-2.0 **score 59.3** ±0.7 · #214 of 732 · from 27 benchmark results across 2 sources Score within the [LMArena (text)](https://whatmodel.sh/?set=lmarena-text&score=set) comparison set only. ## What it's good at _{● marks Qwen3-VL 30B-A3B Thinking among every model on that task or test, worst to best.}

task                vs the others          rank  counts  tests
agent tasks         ▁▁▃▆●▄▅▅▄▄▂▁   #194 of 313*     10%      2
coding              ▁▁▁▃▅▇█●▅▄▃▁   #193 of 469*    4.0%      1
chat & writing      ▁▁▁▃▆▄▆●▆▇▇▃   #216 of 402*    2.5%      1
math                ▁▁▃▅█▆▄●▄▂▁▁   #127 of 431*    8.4%      2
vision & documents  ▁▁▂▂▁▃▄●▆▇█▄     #98 of 153     43%     13

agent tasks · #194 of 313 · 10% of its score

Beats [o1 (2024-12-17)](o1-2024-12-17.md), [o3-mini](o3-mini.md), [DeepSeek R1 0528](deepseek-r1-0528.md) here, though they rank higher overall. Trails [SmolLM3 3B](smollm3-3b.md), [Jamba 1.5 Mini](jamba-1-5-mini.md), [Mellum2 12B-A2.5B Instruct](mellum2-12b-a2-5b-instruct.md) here, though they rank lower overall.

test                      result  vs the others in its run  difficulty            counts  why
τ²-bench Telecom (maker)   27.2%  ▃·●········█ beats 3/6    very hard · avg 9.8%    8.0%  sharp · at its level
BFCL v3 (maker)            68.6%  █▄··▄█·█▄●▄█ beats 10/13  medium · avg 59%        1.9%  blunt · at its level

coding · #193 of 469 · 4.0% of its score

Beats [Gemini 2.5 Flash](gemini-2-5-flash.md), [Devstral 2 (2512)](devstral-2.md), [Qwen3 30B-A3B Thinking 2507](qwen3-30b-a3b-thinking-2507.md) here, though they rank higher overall. Trails [QwQ-32B-Preview](qwq-32b-preview.md), [Ministral 3 8B Reasoning 2512](ministral-3-8b-reasoning.md), [OpenThinker 32B](openthinker-32b.md) here, though they rank lower overall.

test                                result  vs the others in its run  difficulty      counts  why
LiveCodeBench v6 (window unspecif…   64.2%  ▃▅█▅▅▃●···▃█ beats 11/16  hard · avg 35%    4.0%  at its level

chat & writing · #216 of 402 · 2.5% of its score

Beats [Gemma 2 9B IT SimPO](gemma-2-9b-simpo.md), [Granite 4.2 8B](granite-4-2-8b.md), [Mistral Small 4 (119B-A6.5B)](mistral-small-4.md) here, though they rank higher overall. Trails [LFM2.5 1.2B Instruct](lfm2-5-1-2b-instruct.md), [Granite 4.0 H Micro 3B](granite-4-0-h-micro.md), [Gemma 3n E4B](gemma-3n-e4b.md) here, though they rank lower overall.

test                                result  vs the others in its run  difficulty      counts  why
IFEval (aggregation unspecified) …   81.7%  ▃··▃···●█▅▃▃ beats 2/11   easy · avg 77%    2.5%  typical

math · #127 of 431 · 8.4% of its score

Beats [Claude Sonnet 4.6](claude-sonnet-4-6.md), [MiniMax-M3](minimax-m3.md), [MiMo-V2.5 (310B-A15B)](mimo-v2-5.md) here, though they rank higher overall. Trails [Olmo 3.1 32B Think](olmo-3-1-32b-think.md) here, though they rank lower overall.

test                          result  vs the others in its run  difficulty        counts  why
MathVista (testmini) (maker)   81.9%  ▃··▃▅·▃█▅▅●█ beats 13/16  medium · avg 69%    4.4%  typical
AIME 2025 (maker)              83.1%  ▃·▃▃█··▃▅▃·● beats 11/12  medium · avg 44%    3.9%  sharp · at its level

vision & documents · #98 of 153 · 43% of its score

Beats [DeepSeek V4 Flash Vision (Exp)](deepseek-v4-flash-vision-exp.md), [GPT-5 nano](gpt-5-nano.md) here, though they rank higher overall. Trails [GPT-4o (2024-11-20)](gpt-4o-2024-11-20.md) here, though they rank lower overall.

test                                result  vs the others in its run  difficulty           counts  why
CharXiv (reasoning questions) (ma…   56.6%  ▄▂▄▆▄●▆··▄█▂ beats 10/21  medium · avg 38%       9.1%  at its level
MMMU (val) (maker)                   76.0%  ▂···▆▂▂▂▄▄▄● beats 14/16  medium · avg 61%       6.9%  at its level
CharXiv descriptive questions (ma…   86.9%  ▄··▂▂▂·▂▄●▂█ beats 9/14   easy · avg 74%         5.1%  typical
MMStar (maker)                       75.5%  ▂··▄▂▂·▄█▂●▆ beats 13/16  medium · avg 65%       4.6%  at its level
MathVista (testmini) (maker)         81.9%  ▃··▃▅·▃█▅▅●█ beats 13/16  medium · avg 69%       4.4%  typical
MMMU Pro (maker)                     63.0%  ▃·█·▅▃▃▃▅▅●█ beats 13/16  medium · avg 46%       4.2%  at its level
HallusionBench (maker)               66.0%  ▃·▃·▅▃▅▅▃▅█● beats 15/16  medium · avg 54%       1.9%  blunt · at its level
ChartQA (maker)                      89.4%  ▂··▂·▂▃·▅▂▆● beats 13/17  easy · avg 85%         1.4%  blunt
RealWorldQA (maker)                  77.4%  ▅▃·▃▅▅·█·▃●█ beats 12/16  easy · avg 70%         1.3%  blunt
ScreenSpot-Pro (maker)               57.3%  ▂····▂▄··▂●▆ beats 6/11   medium · avg 46%       1.2%  blunt · at its level
DocVQA (maker)                       95.5%  ▃··▅▃▅▃●▃█▅▅ beats 9/17   very easy · avg 94%    1.2%  blunt · maxed out
CC-OCR (maker)                       77.8%  ▃··▃▅·▅●·█▅▃ beats 6/13   easy · avg 74%         0.9%  blunt
OCRBench (maker)                     83.9%  ▅█·●▃█▃▅▅▃·▃ beats 5/16   easy · avg 84%         0.5%  blunt

other tests (9)

test                                result  vs the others in its run  difficulty        counts  why
GPQA (subset unspecified) (maker)    74.4%  ▂····▂·▂▂█▆● beats 13/14  hard · avg 49%      8.1%  sharp · at its level
MMLU-Pro (maker)                     80.5%  ▄·▄·▄▄▄▄█▄█● beats 11/12  medium · avg 63%    6.0%  typical
WebCompass (overall)                 29.6%  ●·▃▃▃···▃·▃█ beats 1/9    hard · avg 18%      4.8%  at its level
MMLU-Redux (maker)                   90.9%  ▃··▃·▃·█▃▅▃● beats 10/11  easy · avg 79%      4.4%  typical
SuperGPQA (maker)                    56.4%  ▄··▄·▄▄▄█▄█● beats 10/11  medium · avg 37%    4.0%  at its level
AndroidWorld (maker)                 55.0%  ▃··▃▅▃▃▃●▃·█ beats 7/11   medium · avg 43%    2.9%  at its level
LVBench (maker)                      59.2%  ▅··▃·▅▃●·█·▃ beats 6/10   hard · avg 51%      2.9%  at its level
Video-MME (without subtitles) (ma…   73.3%  ▃··█··▅█●▃▅▃ beats 9/13   medium · avg 66%    2.6%  blunt · at its level
MVBench (maker)                      72.0%  ▂▂▂·▂█··●▄·▂ beats 8/13   medium · avg 67%    1.1%  blunt · at its level

## How the score was worked out Each result implies a score on its own, after adjusting for how hard that test is and who ran it. The model's score is their weighted average, with a small pull toward the middle for thinly tested models. Each test's weight is its *counts* column above; here it is added up by who ran the tests. [Method](../method.md)

source                                    results  share of the score
Qwen (submitted / maker-reported)              26   95% ▇▇▇▇▇▇▇▇▇▇▇
WebCompass: Towards Multimodal Web Codi…        1    5% ▇
prior: pull toward the middle                        0% ▇
weighted average                                   = 59.3

_{Licences: WebCompass: Towards Multimodal Web Coding Evaluation for Code Language Models ([CC BY-SA 4.0](https://creativecommons.org/licenses/by-sa/4.0/))} ## How to run it ### Through an API api (Alibaba Cloud Model Studio) · $0.75/M blended · $0.2 in / $2.4 out per 1M tokens · [alibabacloud.com](https://www.alibabacloud.com/help/en/model-studio/model-pricing) _{[price source](https://www.alibabacloud.com/help/en/model-studio/model-pricing): USD per 1M tokens, Alibaba Cloud Model Studio, Singapore (International deployment scope), API model qwen3-vl-30b-a3b-thinking (Thinking mode only): input $0.2, output $2.4. Chinese-mainland/Global (US) deployments priced differently and not used, following the International convention of other Alibaba rows; free quota, batch and context-cache discounts excluded. Observed 2026-10-09. · [context source](https://huggingface.co/Qwen/Qwen3-VL-30B-A3B-Thinking/blob/d0ed0380729be07a546fdefafbb4fe411f341e92/README.md): Official Qwen model card: 'Native 256K context, expandable to 1M'. Observed 2026-10-09.} ### llama.cpp (recommended) 1. Install llama.cpp. Homebrew and winget installs name the commands llama-cli and llama-server instead. [Other ways to install](https://github.com/ggml-org/llama.cpp/blob/master/docs/install.md) ```sh # macOS / Linux curl -LsSf https://llama.app/install.sh | sh # Windows irm https://llama.app/install.ps1 | iex ``` 2. Run Qwen3-VL 30B-A3B Thinking (Q4_K_M, needs ~18.6 GB, from [huggingface.co/Qwen](https://huggingface.co/Qwen/Qwen3-VL-30B-A3B-Thinking-GGUF/tree/681477a4711f4bcefa0233abeb16a02abc74c14a)) ```sh llama cli -hf Qwen/Qwen3-VL-30B-A3B-Thinking-GGUF:Q4_K_M ``` 3. Or start a local server and open the address it prints (http://localhost:9931) ```sh llama serve -hf Qwen/Qwen3-VL-30B-A3B-Thinking-GGUF:Q4_K_M ``` Other versions: Q8_0 (~32.5 GB) `llama cli -hf Qwen/Qwen3-VL-30B-A3B-Thinking-GGUF:Q8_0` ### LM Studio 1. Install LM Studio. Open the app once before using the lms command. [Download it (macOS / Windows / Linux)](https://lmstudio.ai/download) 2. Run Qwen3-VL 30B-A3B Thinking (Q4_K_M, needs ~18.6 GB, from [huggingface.co/Qwen](https://huggingface.co/Qwen/Qwen3-VL-30B-A3B-Thinking-GGUF/tree/681477a4711f4bcefa0233abeb16a02abc74c14a)) ```sh lms get https://huggingface.co/Qwen/Qwen3-VL-30B-A3B-Thinking-GGUF@Q4_K_M -y ``` 3. Then run lms ls to see its name, and lms chat followed by that name Other versions: Q8_0 (~32.5 GB) `lms get https://huggingface.co/Qwen/Qwen3-VL-30B-A3B-Thinking-GGUF@Q8_0 -y` · 4bit (~18.3 GB, Mac only) `lms get https://huggingface.co/lmstudio-community/Qwen3-VL-30B-A3B-Thinking-MLX-4bit -y` · 8bit (~33.5 GB, Mac only) `lms get https://huggingface.co/lmstudio-community/Qwen3-VL-30B-A3B-Thinking-MLX-8bit -y` ### Ollama 1. Install Ollama. [Download it (macOS / Windows)](https://ollama.com/download) ```sh # Linux curl -fsSL https://ollama.com/install.sh | sh ``` 2. Run Qwen3-VL 30B-A3B Thinking (Q4_K_M, needs ~20 GB, from [ollama.com](https://ollama.com/library/qwen3-vl/tags)) ```sh ollama run qwen3-vl:30b-a3b-thinking ``` Other versions: Q8_0 (~34 GB) `ollama run qwen3-vl:30b-a3b-thinking-q8_0` ### MLX (Mac) 1. Install MLX (Mac). Apple silicon Macs only, text only. ```sh # with uv uv tool install mlx-lm # or pip install mlx-lm ``` 2. Run Qwen3-VL 30B-A3B Thinking (4bit, needs ~18.3 GB, Mac only, from [huggingface.co/lmstudio-community](https://huggingface.co/lmstudio-community/Qwen3-VL-30B-A3B-Thinking-MLX-4bit/tree/f779740e5e15e7421ff1754b0d1becada8def160)) ```sh mlx_lm.chat --model lmstudio-community/Qwen3-VL-30B-A3B-Thinking-MLX-4bit ``` 3. Or start an OpenAI-compatible server at http://localhost:8080/v1 ```sh mlx_lm.server --model lmstudio-community/Qwen3-VL-30B-A3B-Thinking-MLX-4bit ``` Other versions: 8bit (~33.5 GB, Mac only) `mlx_lm.chat --model lmstudio-community/Qwen3-VL-30B-A3B-Thinking-MLX-8bit` Model repository and documentation: [huggingface.co/Qwen](https://huggingface.co/Qwen/Qwen3-VL-30B-A3B-Thinking) [interactive version](https://whatmodel.sh/model/qwen3-vl-30b-a3b-thinking) · [JSON](https://whatmodel.sh/api?format=json&all=1)