whatmodel.sh / Qwen3-VL 30B-A3B Thinking

Qwen3-VL 30B-A3B Thinking

+ contribute
Update model informationAdd a place to run itAdd a test resultPropose a benchmarkCorrect a benchmark scoreEdit data on GitHub

Qwen3-VL 30B-A3B Thinking is an open-weights language model from Alibaba, released September 2025. Its published size is 31.1B parameters (3B active per token) and a 256k-token context window. Licence: apache-2.0.

Catalogue identity notes

Maker repo Qwen/Qwen3-VL-30B-A3B-Thinking @d0ed038 (licence tag apache-2.0, created 2025-09-30); the separate Thinking checkpoint, not Qwen3-VL-30B-A3B-Instruct. Scored from the Qwen3-VL Technical Report (arXiv 2511.21631v2, Tables 3/4 and 7-10: column 'Qwen3-VL 30B-A3B thinking'), whose instruct and thinking columns are the two published checkpoints. Released 2025-09: month the maker's Hugging Face repo was created.

Model information sources

Release date ↗ · Weights licence ↗

31,070,754,032 parameters in the Hugging Face safetensors metadata of Qwen/Qwen3-VL-30B-A3B-Thinking @d0ed038 (vision encoder included); MoE with ~3B activated per token per the model name ('30B-A3B').

59.7±0.8#200 of 719 LLMsfrom 26 benchmark results across 1 source

$ how it compares

The LLMs ranked either side of Qwen3-VL 30B-A3B Thinking, out of 828 with a score, then more from Alibaba.

#modelscorewhereRAMreleased
239Granite 4.2 8B IBM60±1api $0.11/M +local~6.6GB2026-08
240Solar Open 100B (102B-A12B) Upstage60±1local~41GB2025-12
241Qwen3-VL 30B-A3B Thinking Alibaba60±1api $0.75/M +local~21GB2025-09
242Mistral Small 4 (119B-A6.5B) Mistral60±2api $0.26/M +local~40GB2026-03
243Gemini 2.0 Flash-Lite Preview (02-05) Google60±3comparison only2025-02
more from Alibaba
18Qwen3.8 Max (2.4T-A95B) Alibaba78±1api $3.00/M +local~380GB2026-08
34Qwen3.8 Max (0902) Alibaba76±1api $3.00/M2026-09
37Qwen3.8-Flash-Next Alibaba75±1local~59GB2026-08

see all LLMs · find the ones that fit your machine

$ what it's good at

agent tasks▁▁▃▆█▆▅▅▅▄▃▁ #182 of 300 * · 10% of its score · 2 tests

Beats o1 (2024-12-17), o3-mini, DeepSeek R1 0528 here, though they rank higher overall. Trails SmolLM3 3B, Jamba 1.5 Mini, Mellum2 12B-A2.5B Instruct here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
τ²-bench Telecom lab27.2% ✎▃·█········█ beats 3/6very hard · avg 10%▇▇▇ 8.2%sharp · at its level
BFCL v3 lab68.6% ✎█▄··▄█·█▄█▄█ beats 10/13medium · avg 59%▇ 1.9%blunt · at its level
coding▁▁▁▃▆██▇▆▄▃▁ #180 of 458 * · 4.3% of its score · 1 test

Beats Qwen3.5 9B, Gemini 2.5 Flash, Qwen3 30B-A3B Thinking 2507 here, though they rank higher overall. Trails QwQ-32B-Preview, OpenThinker 32B, Ministral 3 8B Reasoning 2512 here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
LiveCodeBench v6 (window unspecified) lab64.2% ✎▃▅█▅▅▃▅···▃█ beats 11/16medium · avg 35%▇▇ 4.3%at its level
chat & writing▁▁▁▃▅▄▅█▆▆▇▃ #209 of 393 * · 2.5% of its score · 1 test

Beats Granite 4.2 8B here, though they rank higher overall. Trails LFM2.5 1.2B Instruct, Granite 4.0 H Micro 3B, Gemma 3n E4B here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
IFEval (aggregation unspecified) lab81.7% ✎▃··▃···██▅▃▃ beats 2/11easy · avg 77%▇ 2.5%typical
math▁▂▃▅▆▅▄█▄▁▁▁ #126 of 398 * · 8.9% of its score · 2 tests

Beats Claude Sonnet 4.6, MiniMax-M3, MiMo-V2.5 (310B-A15B) here, though they rank higher overall. Trails Olmo 3.1 32B Think here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
MathVista (testmini) lab81.9% ✎▃··▃▅·▃█▅▅▅█ beats 13/16medium · avg 69%▇▇ 4.8%typical
AIME 2025 lab83.1% ✎▃·▃▃█··▃▅▃·█ beats 11/12medium · avg 45%▇▇ 4.1%sharp · at its level
vision & documents▁▁▃▂▁▃▄▃▆▇█▄ #91 of 147 · 45% of its score · 13 tests

Beats DeepSeek V4 Flash Vision (Exp), GPT-5 nano here, though they rank higher overall. Trails Mistral Small 4 (119B-A6.5B) here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
CharXiv (reasoning questions) lab56.6% ✎▄▂▄▆▄▄▆··▄█▂ beats 10/21medium · avg 39%▇▇▇▇ 9.2%at its level
MMMU (val) lab76.0% ✎▂···▆▂▂▂▄▄▄█ beats 14/16medium · avg 61%▇▇▇ 7.4%at its level
CharXiv descriptive questions lab86.9% ✎▄··▂▂▂·▂▄▄▂█ beats 9/14easy · avg 74%▇▇ 5.2%typical
MMStar lab75.5% ✎▂··▄▂▂·▄█▂▄▆ beats 13/16medium · avg 65%▇▇ 4.9%at its level
MathVista (testmini) lab81.9% ✎▃··▃▅·▃█▅▅▅█ beats 13/16medium · avg 69%▇▇ 4.8%typical
MMMU Pro lab63.0% ✎▃·█·▅▃▃▃▅▅▃█ beats 13/16medium · avg 46%▇▇ 4.5%at its level
HallusionBench lab66.0% ✎▃·▃·▅▃▅▅▃▅█▅ beats 15/16medium · avg 55%▇ 2.0%blunt · at its level
ChartQA lab89.4% ✎▂··▂·▂▃·▅▂▆█ beats 13/17easy · avg 85%▇ 1.4%blunt
5 more tests
testresultvs the others in its rundifficultycountswhy
RealWorldQA lab77.4% ✎▅▃·▃▅▅·█·▃▅█ beats 12/16easy · avg 70%▇ 1.4%blunt
ScreenSpot-Pro lab57.3% ✎▂····▂▄··▂█▆ beats 6/11medium · avg 46%▇ 1.2%blunt · at its level
DocVQA lab95.5% ✎▃··▅▃▅▃█▃█▅▅ beats 9/17very easy · avg 94%▇ 1.2%blunt · maxed out
CC-OCR lab77.8% ✎▃··▃▅·▅▅·█▅▃ beats 6/13easy · avg 74%▇ 0.9%blunt
OCRBench lab83.9% ✎▅█·▃▃█▃▅▅▃·▃ beats 5/16easy · avg 84%▇ 0.5%blunt
other tests8 not in a task · 34% of its score
testresultvs the others in its rundifficultycountswhy
GPQA (subset unspecified) lab74.4% ✎▂····▂·▂▂█▆█ beats 13/14hard · avg 49%▇▇▇ 8.6%sharp · at its level
MMLU-Pro lab80.5% ✎▄·▄·▄▄▄▄█▄██ beats 11/12medium · avg 63%▇▇ 6.1%typical
MMLU-Redux lab90.9% ✎▃··▃·▃·█▃▅▃▅ beats 10/11easy · avg 79%▇▇ 4.7%typical
SuperGPQA lab56.4% ✎▄··▄·▄▄▄█▄██ beats 10/11medium · avg 37%▇▇ 4.4%at its level
AndroidWorld lab55.0% ✎▃··▃▅▃▃▃▃▃·█ beats 7/11medium · avg 43%▇ 3.1%blunt · at its level
LVBench lab59.2% ✎▅··▃·▅▃▃·█·▃ beats 6/10hard · avg 51%▇ 3.1%at its level
Video-MME (without subtitles) lab73.3% ✎▃··█··▅█▃▃▅▃ beats 9/13medium · avg 66%▇ 2.7%blunt · at its level
MVBench lab72.0% ✎▂▂▂·▂█··▆▄·▂ beats 8/13medium · avg 67%▇ 1.1%blunt · at its level

$ where to run it

Through an API

api · Alibaba Cloud Model Studio · $0.75/M · $0.2 in / $2.4 out per 1M tokens · 256k context · www.alibabacloud.com ↗ · price sourceUSD 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.

llama.cpp · recommended

version

  1. Install llama.cpp · Homebrew and winget installs name the commands llama-cli and llama-server instead.
    macOS / Linux
    curl -LsSf https://llama.app/install.sh | sh
    Windows
    irm https://llama.app/install.ps1 | iex

    other ways to install ↗

  2. Run Qwen3-VL 30B-A3B Thinking · needs ~18.6 GB · from huggingface.co/Qwen (official)
    chat in the terminal
    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)
    llama serve -hf Qwen/Qwen3-VL-30B-A3B-Thinking-GGUF:Q4_K_M

LM Studio

version

  1. Install LM Studio · Open the app once before using the lms command.

    download it (macOS / Windows / Linux) ↗

  2. Run Qwen3-VL 30B-A3B Thinking · needs ~18.6 GB · from huggingface.co/Qwen (official)

    open in LM Studio ↗

  3. Or download it in the terminal
    lms get https://huggingface.co/Qwen/Qwen3-VL-30B-A3B-Thinking-GGUF@Q4_K_M -y
  4. Then run lms ls to see its name, and lms chat followed by that name

Ollama

version

  1. Install Ollama

    download it (macOS / Windows) ↗

    Linux
    curl -fsSL https://ollama.com/install.sh | sh
  2. Run Qwen3-VL 30B-A3B Thinking · needs ~20 GB · from ollama.com
    chat in the terminal
    ollama run qwen3-vl:30b-a3b-thinking

MLX (Mac)

version

  1. Install MLX (Mac) · Apple silicon Macs only, text only.
    with uv
    uv tool install mlx-lm
    or
    pip install mlx-lm
  2. Run Qwen3-VL 30B-A3B Thinking · needs ~18.3 GB, Mac only · from huggingface.co/lmstudio-community
    chat in the terminal
    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
    mlx_lm.server --model lmstudio-community/Qwen3-VL-30B-A3B-Thinking-MLX-4bit

Official weights and documentation: huggingface.co/Qwen ↗

$ how the score was worked out · Qwen (Alibaba) vendor-reporte… 100%

Each result implies a score on its own, after adjusting for how hard its test is and who ran it. The 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's added up by who ran the tests. How it works

sourceresultsshare of the scorelicence
Qwen (Alibaba) vendor-reported benchmark scores for its own models (qwen.ai launch posts, Qwen2.5-VL and Qwen3-VL technical reports, Qwen3.8-Flash-Next card) · lab-reported26▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 100%Factual vendor-reported score (not licensed content)
prior · pull toward the middle▇ 0%
weighted average= 59.7

$ improve this page

Update model informationAdd a place to run itAdd a test resultPropose a benchmarkCorrect a benchmark scoreEdit data on GitHub

Use a short GitHub form or edit the data through a pull request. Every number needs a source; here's which ones we can use.

data updated · sources and licences · json