[whatmodel.sh](../../README.md) / Qwen3-VL 32B Thinking
# Qwen3-VL 32B Thinking
Alibaba · released 2025-10 · 33.4B params · apache-2.0
**score 61.6** ±0.7 · #186 of 732 · from 35 benchmark results across 3 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 32B Thinking among every model on that task or test, worst to best.
task vs the others rank counts tests
agent tasks ▁▁▃▆█●▅▅▄▄▂▁ #166 of 313 21% 5
coding ▁▁▁▃▅▇█●▅▄▃▁ #184 of 469* 3.6% 1
chat & writing ▁▁▁▃▆▄▆█●▇▇▃ #158 of 402* 2.1% 1
math ▁▁▃▅█▆▄●▄▂▁▁ #111 of 431 8.7% 4
vision & documents ▁▁▂▂▁▃▄▂●▇█▄ #89 of 153 38% 13
agent tasks · #166 of 313 · 21% 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), [Jamba 1.5 Large](jamba-1-5-large.md) here, though they rank lower overall.
test result vs the others in its run difficulty counts why
τ²-bench Telecom (maker) 46.9% ▃·█········● beats 6/6 very hard · avg 9.8% 11% sharp · at its level · several runs
Terminal-Bench Hard 6.9% ▂▅▆▂●▆▂▂█▅▅▂ beats 9/27 very hard · avg 2.1% 3.5% sharp · small run
τ²-bench Retail 57.0% ▁····▂▁●▆█▃▁ beats 7/44 medium · avg 37% 2.5% at its level
τ²-bench Airline 54.4% ▁▁·▂▁▃▂●█▅▂▃ beats 15/42 hard · avg 29% 2.0% at its level
BFCL v3 (maker) 71.7% █▄··▄█·█▄█▄● beats 12/13 medium · avg 59% 1.8% blunt · at its level
coding · #184 of 469 · 3.6% of its score
Beats [Qwen3.5 9B](qwen3-5-9b.md), [Gemini 2.5 Flash](gemini-2-5-flash.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), [Phi-4-reasoning 14B](phi-4-reasoning.md) here, though they rank lower overall.
test result vs the others in its run difficulty counts why
LiveCodeBench v6 (window unspecif… 65.6% ▃▅█▅▅▃●···▃█ beats 12/16 hard · avg 35% 3.6% at its level
chat & writing · #158 of 402 · 2.1% of its score
Beats [INTELLECT-3](intellect-3.md), [MiniMax-M2](minimax-m2.md), [MiniCPM5 2B](minicpm5-2b.md) here, though they rank higher overall. Trails [LFM2.5 1.2B Thinking](lfm2-5-1-2b-thinking.md), [LFM2.5 8B-A1B](lfm2-5-8b-a1b.md), [LFM2.5 2.6B](lfm2-5-2-6b.md) here, though they rank lower overall.
test result vs the others in its run difficulty counts why
IFEval (aggregation unspecified) … 87.8% ▃··▃···██▅▃● beats 11/11 easy · avg 77% 2.1% typical
math · #111 of 431 · 8.7% of its score
Beats [Claude Sonnet 4.6](claude-sonnet-4-6.md), [MiniMax-M3](minimax-m3.md), [Nemotron 3 Ultra 550B-A55B](nemotron-3-ultra-550b-a55b.md) here, though they rank higher overall.
test result vs the others in its run difficulty counts why
MathVista (testmini) (maker) 85.9% ▃··▃▅·▃█▅▅▅● beats 16/16 medium · avg 69% 3.8% typical
AIME 2025 (maker) 83.7% ▃·▃▃█··▃▅▃·● beats 12/12 medium · avg 44% 3.2% sharp
OMEGA (math generalisation) 50.8% ▅···▃▃···▃▃● beats 6/8 hard · avg 24% 1.3% at its level
MATH (subset unspecified) 96.7% ▅····▃▃··▃▃● beats 8/8 easy · avg 73% 0.5% noisy
vision & documents · #89 of 153 · 38% of its score
Beats [DeepSeek V4 Flash Vision (Exp)](deepseek-v4-flash-vision-exp.md), [GPT-5 mini](gpt-5-mini.md), [GLM-4.6V](glm-4-6v.md) here, though they rank higher overall. Trails [GPT-4.1 mini](gpt-4-1-mini.md), [Mistral Large 3 (675B-A41B)](mistral-large-3.md) here, though they rank lower overall.
test result vs the others in its run difficulty counts why
CharXiv (reasoning questions) (ma… 65.2% ▄▂▄▆▄▄●··▄█▂ beats 13/21 medium · avg 38% 8.3% at its level
MMMU (val) (maker) 78.1% ▂···▆▂▂▂▄▄▄● beats 15/16 medium · avg 61% 6.0% at its level
CharXiv descriptive questions (ma… 90.2% ▄··▂▂▂·▂▄▄▂● beats 12/14 easy · avg 74% 4.1% typical
MMStar (maker) 79.4% ▂··▄▂▂·▄█▂▄● beats 16/16 medium · avg 65% 4.1% at its level
MMMU Pro (maker) 68.1% ▃·█·▅▃▃▃▅▅▃● beats 15/16 medium · avg 46% 3.9% at its level
MathVista (testmini) (maker) 85.9% ▃··▃▅·▃█▅▅▅● beats 16/16 medium · avg 69% 3.8% typical
HallusionBench (maker) 67.4% ▃·▃·▅▃▅▅▃▅█● beats 16/16 medium · avg 54% 1.8% blunt · at its level
RealWorldQA (maker) 78.4% ▅▃·▃▅▅·█·▃▅● beats 14/16 easy · avg 70% 1.2% blunt
ChartQA (maker) 89.0% ▂··▂·▂▃·▅▂●█ beats 12/17 easy · avg 85% 1.2% blunt
ScreenSpot-Pro (maker) 57.1% ▂····▂▄··▂●▆ beats 5/11 medium · avg 46% 1.1% blunt · at its level
DocVQA (maker) 96.1% ▃··▅▃▅▃█▃●▅▅ beats 12/17 very easy · avg 94% 1.0% blunt · maxed out
CC-OCR (maker) 79.6% ▃··▃▅·▅▅·●▅▃ beats 8/13 easy · avg 74% 0.8% blunt
OCRBench (maker) 85.5% ▅█·▃●█▃▅▅▃·▃ beats 6/16 easy · avg 84% 0.4% blunt
other tests (11)
test result vs the others in its run difficulty counts why
GPQA (subset unspecified) (maker) 73.1% ▂····▂·▂▂█▆● beats 11/14 hard · avg 49% 7.1% sharp · at its level
MMLU-Pro (maker) 82.1% ▄·▄·▄▄▄▄█▄█● beats 12/12 medium · avg 63% 5.0% typical
SuperGPQA (maker) 59.0% ▄··▄·▄▄▄█▄█● beats 11/11 medium · avg 37% 3.7% at its level
MMLU-Redux (maker) 91.9% ▃··▃·▃·█▃▅▃● beats 11/11 easy · avg 79% 3.6% typical
AndroidWorld (maker) 63.7% ▃··▃▅▃▃▃▃▃·● beats 10/11 medium · avg 43% 2.7% at its level
LVBench (maker) 62.6% ▅··▃·▅▃▃·●·▃ beats 8/10 hard · avg 51% 2.7% at its level
Video-MME (without subtitles) (ma… 77.3% ▃··█··▅█▃▃●▃ beats 12/13 medium · avg 66% 2.3% blunt · at its level
MVBench (maker) 73.2% ▂▂▂·▂█··▆●·▂ beats 11/13 medium · avg 67% 1.0% blunt · at its level
HellaSwag 62.1% ▁▁▁▂▂▂▄●█▇▄▂ beats 44/135 medium · avg 63% 1.0% at its level · small run
MMLU 71.5% ▁▁▁▁▁▂▂▅█●▆▄ beats 145/214 medium · avg 68% 0.9% noisy · small run
PIQA 88.0% ▁▁▁▂▂▃▅█▅▃▁● beats 122/123 easy · avg 83% 0.2% blunt · small run
source results share of the score
Qwen (submitted / maker-reported) 26 85% ▇▇▇▇▇▇▇▇▇▇
HyperCLOVA X 32B Think 7 13% ▇▇
allenai/Olmo-3.1-32B-Think model card 2 2% ▇
prior: pull toward the middle 0% ▇
weighted average = 61.6
Licences: HyperCLOVA X 32B Think ([CC-BY-4.0](https://creativecommons.org/licenses/by/4.0/)) · allenai/Olmo-3.1-32B-Think model card ([Apache-2.0](../../licenses/allenai--Olmo-3.1-32B-Think.txt))
## How to run it
### Through an API
api (Alibaba Cloud Model Studio) · $0.28/M blended · $0.16 in / $0.64 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, API model qwen3-vl-32b-thinking ('Thinking mode only', deployment scope 'Global'): input $0.16, output $0.64, the same in both regional tables that list it and equal to the catalogued qwen3-vl-32b-instruct row. Free quota, batch and context-cache discounts excluded. Observed 2026-10-11. · [context source](https://huggingface.co/Qwen/Qwen3-VL-32B-Thinking/blob/7edd10ffd1196091948fb245ff63e406ccb2d4d1/README.md): Context length stated on the official model card.
### 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 32B Thinking (Q4_K_M, needs ~19.8 GB, from [huggingface.co/Qwen](https://huggingface.co/Qwen/Qwen3-VL-32B-Thinking-GGUF/tree/a612b3102374e69a892162730ad904b69961f806))
```sh
llama cli -hf Qwen/Qwen3-VL-32B-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-32B-Thinking-GGUF:Q4_K_M
```
Other versions: Q8_0 (~34.8 GB) `llama cli -hf Qwen/Qwen3-VL-32B-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 32B Thinking (Q4_K_M, needs ~19.8 GB, from [huggingface.co/Qwen](https://huggingface.co/Qwen/Qwen3-VL-32B-Thinking-GGUF/tree/a612b3102374e69a892162730ad904b69961f806))
```sh
lms get https://huggingface.co/Qwen/Qwen3-VL-32B-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 (~34.8 GB) `lms get https://huggingface.co/Qwen/Qwen3-VL-32B-Thinking-GGUF@Q8_0 -y` · 4bit (~19.6 GB, Mac only) `lms get https://huggingface.co/lmstudio-community/Qwen3-VL-32B-Thinking-MLX-4bit -y` · 8bit (~36 GB, Mac only) `lms get https://huggingface.co/lmstudio-community/Qwen3-VL-32B-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 32B Thinking (Q4_K_M, needs ~21 GB, from [ollama.com](https://ollama.com/library/qwen3-vl/tags))
```sh
ollama run qwen3-vl:32b-thinking
```
Other versions: Q8_0 (~36 GB) `ollama run qwen3-vl:32b-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 32B Thinking (4bit, needs ~19.6 GB, Mac only, from [huggingface.co/lmstudio-community](https://huggingface.co/lmstudio-community/Qwen3-VL-32B-Thinking-MLX-4bit/tree/13ddf0e8f54ebeddbc7790fd15e272b33c5f6568))
```sh
mlx_lm.chat --model lmstudio-community/Qwen3-VL-32B-Thinking-MLX-4bit
```
3. Or start an OpenAI-compatible server at http://localhost:8080/v1
```sh
mlx_lm.server --model lmstudio-community/Qwen3-VL-32B-Thinking-MLX-4bit
```
Other versions: 8bit (~36 GB, Mac only) `mlx_lm.chat --model lmstudio-community/Qwen3-VL-32B-Thinking-MLX-8bit`
Model repository and documentation: [huggingface.co/Qwen](https://huggingface.co/Qwen/Qwen3-VL-32B-Thinking)
[interactive version](https://whatmodel.sh/model/qwen3-vl-32b-thinking) · [JSON](https://whatmodel.sh/api?format=json&all=1)