whatmodel.sh / Qwen3-VL 235B-A22B Thinking

Qwen3-VL 235B-A22B Thinking

+ contribute

Qwen3-VL 235B-A22B Thinking is an open-weights language model from Alibaba, released September 2025. Its published size is 235B parameters (22B active per token) and a 256k-token context window. Licence: apache-2.0.

Catalogue identity notes

Arena entry 'qwen3-vl-235b-a22b-thinking' (text, vision). Maker repo Qwen/Qwen3-VL-235B-A22B-Thinking @6664aff: licence tag apache-2.0. Parameters: 235B total / 22B active per the model name; HF safetensors metadata 235.7B. Released 2025-09: month the maker's Hugging Face repo was created (createdAt 2025-09-22). Separate Thinking checkpoint (not the Instruct weights).

64.1±0.8#156 of 719 LLMsfrom 50 benchmark results across 4 sources

$ how it compares

The LLMs ranked either side of Qwen3-VL 235B-A22B Thinking, out of 828 with a score, then more from Alibaba.

#	model	score	where	RAM	released
176	Claude Haiku 4.5 Anthropic	64±1	api $2.00/M		2025-10
177	Nemotron-Labs-3 Puzzle 75B-A9B NVIDIA	64±1	local	~35GB	2026-06
178	Qwen3-VL 235B-A22B Thinking Alibaba	64±1	local	~111GB	2025-09
179	Qwen3-Coder 480B-A35B Instruct Alibaba	64±1	api $0.67/M +local	~78GB	2025-07
180	Mistral Medium 3.1 (2508) Mistral	64±2	comparison only		2025-08
	more from Alibaba
18	Qwen3.8 Max (2.4T-A95B) Alibaba	78±1	api $3.00/M +local	~380GB	2026-08
34	Qwen3.8 Max (0902) Alibaba	76±1	api $3.00/M		2026-09
37	Qwen3.8-Flash-Next Alibaba	75±1	local	~59GB	2026-08

see all LLMs · find the ones that fit your machine

$ what it's good at

agent tasks▁▁▃▆█▆▅▅▅▄▃▁ #152 of 300 · 13% of its score · 3 tests

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

test	result	vs the others in its run	difficulty	counts	why
τ²-bench Telecom lab	44.7% ✎	▃·█········█ beats 4/6	very hard · avg 10%	▇▇▇▇ 9.2%	sharp · at its level
OSWorld-Verified lab	38.1% ✎	█▄·▂▆▂▄▂▆▄▆█ beats 1/25	hard · avg 20%	▇ 1.9%	at its level · small run
BFCL v3 lab	71.8% ✎	█▄··▄█·█▄█▄█ beats 13/13	medium · avg 59%	▇ 1.8%	blunt · at its level

coding▁▁▁▃▆██▇▆▄▃▁ #130 of 458 * · 1.8% of its score · 1 test

Beats Qwen3.5 Plus, Gemini 2.5 Pro, Gemini 3.1 Flash-Lite here, though they rank higher overall. Trails Devstral Small 2 24B, Llama 3.1 Nemotron Ultra 253B v1, Nemotron 3.5 Lightning 30B-A3B here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
LiveCodeBench v6 (Feb–May 2025) lab	70.1% ✎	▁▁▃▁█▅▄▇▅▅▃█ beats 21/38	medium · avg 41%	▇ 1.8%	at its level · small run

chat & writing▁▁▁▃▅▄▅█▆▆▇▃ #118 of 393 * · 6.2% of its score · 2 tests

Beats LongCat 2.0 (1.6T-A48B), GPT-5 mini, Qwen3.5 35B-A3B here, though they rank higher overall. Trails LFM2.5 2.6B, Nemotron-Cascade-8B, DeepSeek V3 0324 here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
LMArena (text)	1395.0 ✎	▁▁▁▂▃▃▃▆▆▇█▃ beats 239/379	avg 1268	▇▇ 5.1%	blunt
IFEval (aggregation unspecified) lab	88.2% ✎	▁▁▃▁▂▄▄▆█▃▂▁ beats 96/141	easy · avg 77%	▇ 1.1%	small run

math▁▂▃▅▆▅▄█▄▁▁▁ #117 of 398 · 15% of its score · 6 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, Gemini 2.5 Flash, Qwen3-VL 32B Thinking here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
HMMT 2025 (edition unspecified) lab	77.4% ✎	▃█▇▅▇▅▁·▁··▁ beats 22/32	hard · avg 34%	▇▇ 4.0%	sharp · small run
AIME 2025 lab	89.7% ✎	▁·▁▁▁▂▅▆█▇▅▁ beats 115/133	medium · avg 56%	▇▇ 3.8%	sharp · small run
MATH-Vision lab	74.6% ✎	█·····▄▄·▄·▄ beats 1/5	medium · avg 46%	▇ 2.7%	at its level
MathVista (testmini) lab	85.8% ✎	▁··▁▄▄▇█▅▆█▄ beats 40/47	medium · avg 65%	▇ 1.8%	small run
DynaMath lab	82.8% ✎	▂·····▂··█▄▆ beats 3/10	medium · avg 64%	▇ 1.4%	small run
We-Math lab	74.8% ✎	█··········▃ beats 0/3	medium · avg 56%	▇ 1.2%	small run

vision & documents▁▁▃▂▁▃▄▂▆▇█▄ #79 of 147 · 32% of its score · 18 tests

Beats DeepSeek V4 Flash Vision (Exp), GPT-5.4 nano, GPT-5 mini here, though they rank higher overall. Trails GPT-4.1 mini, Gemini 2.5 Flash, GPT-4.1 here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
CharXiv (reasoning questions) lab	66.1% ✎	▄▂▄▆▄▄▆··▄█▂ beats 14/21	medium · avg 39%	▇▇▇ 8.3%	at its level
CharXiv descriptive questions lab	90.5% ✎	▄··▂▂▂·▂▄▄▂█ beats 14/14	easy · avg 74%	▇▇ 3.9%	typical
Arena vision (style controlled)	1188.0 ✎	▁▁▂▂▂▂▄▅▅▆██ beats 64/145	avg 1068	▇ 3.5%	blunt
MATH-Vision lab	74.6% ✎	█·····▄▄·▄·▄ beats 1/5	medium · avg 46%	▇ 2.7%	at its level
MathVista (testmini) lab	85.8% ✎	▁··▁▄▄▇█▅▆█▄ beats 40/47	medium · avg 65%	▇ 1.8%	small run
MMMU (val) lab	80.6% ✎	▁····▁▁▄█▇▆▆ beats 46/52	medium · avg 59%	▇ 1.7%	at its level · small run
MMMU Pro lab	69.3% ✎	▂▃▁▃▆▇▃▇▇██▅ beats 33/52	medium · avg 47%	▇ 1.6%	at its level · small run
MMStar lab	78.7% ✎	▃▅▂▃▂▂▆█▆▆█▅ beats 26/34	medium · avg 63%	▇ 1.3%	at its level · small run

10 more tests

test	result	vs the others in its run	difficulty	counts	why
ChartQA lab	90.3% ✎	▂··▂·▂▃·▅▂▆█ beats 17/17	easy · avg 85%	▇ 1.2%	blunt
ScreenSpot-Pro lab	61.8% ✎	▂····▂▄··▂█▆ beats 10/11	medium · avg 46%	▇ 1.2%	blunt · at its level
CountBench lab	93.7% ✎	▂▂▃▂·▂▂▃▃·▅█ beats 10/18	very easy · avg 89%	▇ 1.1%	maxed out · small run
DocVQA lab	96.5% ✎	▃··▅▃▅▃█▃█▅▅ beats 15/17	very easy · avg 94%	▇ 1.0%	blunt · maxed out
SimpleVQA lab	61.3% ✎	▃·▃·▃·▃·█▅▃▃ beats 8/10	medium · avg 37%	▇ 1.0%	at its level · small run
HallusionBench lab	66.7% ✎	▃▁·▃▃▃▄▅▄█▅▅ beats 24/32	medium · avg 56%	▇ 0.6%	blunt · at its level · small run
RealWorldQA lab	81.3% ✎	▂▃▂█▃██▆█▃▆▆ beats 30/38	easy · avg 72%	▇ 0.5%	blunt · small run
CC-OCR lab	81.5% ✎	▁·▂··▂▂▃▃▂▆█ beats 22/26	easy · avg 72%	▇ 0.4%	blunt · small run
OCRBench lab	87.5% ✎	▁·▃▁▆▅▂█▅█▁▃ beats 23/37	easy · avg 83%	▇ 0.4%	blunt · small run
AI2D test lab	89.2% ✎	▁▁▅▇▃▇·▄·█·▃ beats 18/28	easy · avg 85%	▇ 0.2%	blunt · small run

other tests22 not in a task · 36% of its score

test	result	vs the others in its run	difficulty	counts	why
GPQA (subset unspecified) lab	77.1% ✎	▂····▂·▂▂█▆█ beats 14/14	hard · avg 49%	▇▇▇ 7.3%	sharp · at its level
AndroidWorld lab	62.0% ✎	▃··▃▅▃▃▃▃▃·█ beats 9/11	medium · avg 43%	▇ 3.0%	blunt · at its level
Humanity's Last Exam lab	13.6% ✎	▁▁▁▃▆▆▇██▆▁▃ beats 31/104	very hard · avg 4.5%	▇ 2.2%	small run
MMLU-Pro lab	83.8% ✎	▁▁·▁▁▁▃▂▄▅█▁ beats 160/188	medium · avg 59%	▇ 2.1%	small run
INCLUDE lab	80.0% ✎	▂····▂▂▄▄▄█▄ beats 28/36	medium · avg 57%	▇ 2.0%	at its level · small run
MMLU-Redux lab	93.7% ✎	▁▁▃▁▂▃▄▅▇▅█▄ beats 58/63	easy · avg 81%	▇ 1.9%	small run
Creative Writing v3 lab	85.7% ✎	▂····▂▆··▆█▆ beats 9/14	easy · avg 67%	▇ 1.5%	small run
VideoMMMU lab	80.0% ✎	▂···▂···▂▂█▆ beats 4/10	medium · avg 61%	▇ 1.4%	small run

14 more tests

test	result	vs the others in its run	difficulty	counts	why
MMLU-ProX lab	80.6% ✎	▁·▁▁▁▂▁▃▂█▅▄ beats 24/35	medium · avg 59%	▇ 1.3%	small run
SuperGPQA lab	64.3% ✎	▂▁▂▂▅▆▃▇▇▇█▃ beats 39/46	medium · avg 38%	▇ 1.3%	at its level · small run
WritingBench lab	86.7% ✎	▁·▁▂▁··▅█▂▃▁ beats 18/20	easy · avg 70%	▇ 1.2%	small run
MLVU lab	83.8% ✎	▂···▂···▂▄▄█ beats 3/10	easy · avg 69%	▇ 1.1%	small run
MMVU lab	71.1% ✎	▃···▃···▃▃▅█ beats 4/8	medium · avg 48%	▇ 1.1%	at its level · small run
LVBench lab	63.6% ✎	▇·█▅▅▄▄··▃▁▁ beats 15/28	hard · avg 51%	▇ 1.1%	at its level · small run
RefSpatial-Bench lab	69.9% ✎	▃·▃····▃▃▅▅█ beats 8/10	medium · avg 43%	▇ 1.1%	at its level · small run
LiveBench 2024-11-25 lab	79.6% ✎	▄··▂·▂·▂█▆▂▃ beats 13/13	medium · avg 59%	▇ 1.1%	small run
ERQA lab	52.5% ✎	▃··▃··█·▃█▅▅ beats 2/12	medium · avg 36%	▇ 1.1%	at its level · small run
MMLongBench-Doc lab	56.2% ✎	▃····▃··▃▃▅█ beats 3/8	medium · avg 36%	▇ 1.0%	at its level · small run
Multi-IF (aggregation unspecified) lab	79.1% ✎	▁▁▁▁▁▁▃▄█▄▁▂ beats 29/41	medium · avg 62%	▇ 1.0%	blunt · small run
EmbSpatial-Bench lab	84.3% ✎	▁·····▁·▁·▃█ beats 7/10	easy · avg 68%	▇ 1.0%	small run
Video-MME (without subtitles) lab	79.0% ✎	▁▃▁▃▁·█▄▄▅·▇ beats 21/27	medium · avg 67%	▇ 0.9%	blunt · at its level · small run
MVBench lab	75.2% ✎	▁··▃▄▁▃▄▄▄█▄ beats 22/26	medium · avg 66%	▇ 0.4%	blunt · at its level · small run

$ where to run it

llama.cpp · recommended

version

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 ↗
Run Qwen3-VL 235B-A22B Thinking · needs ~142.2 GB · from huggingface.co/Qwen (official)
chat in the terminal
llama cli -hf Qwen/Qwen3-VL-235B-A22B-Thinking-GGUF:Q4_K_M
Or start a local server and open the address it prints (http://localhost:9931)
llama serve -hf Qwen/Qwen3-VL-235B-A22B-Thinking-GGUF:Q4_K_M

LM Studio

version

Install LM Studio · Open the app once before using the lms command.
download it (macOS / Windows / Linux) ↗
Run Qwen3-VL 235B-A22B Thinking · needs ~142.2 GB · from huggingface.co/Qwen (official)
open in LM Studio ↗
Or download it in the terminal
lms get https://huggingface.co/Qwen/Qwen3-VL-235B-A22B-Thinking-GGUF@Q4_K_M -y
Then run lms ls to see its name, and lms chat followed by that name

Ollama

version

Install Ollama
download it (macOS / Windows) ↗
Linux
curl -fsSL https://ollama.com/install.sh | sh
Run Qwen3-VL 235B-A22B Thinking · needs ~141.5 GB · from ollama.com
chat in the terminal
ollama run qwen3-vl:235b-a22b-thinking-q4_K_M

MLX (Mac)

Install MLX (Mac) · Apple silicon Macs only, text only.
with uv
uv tool install mlx-lm
or
pip install mlx-lm
Run Qwen3-VL 235B-A22B Thinking · needs ~104 GB, Mac only · from huggingface.co/mlx-community
chat in the terminal
mlx_lm.chat --model mlx-community/Qwen3-VL-235B-A22B-Thinking-3bit
Or start an OpenAI-compatible server at http://localhost:8080/v1
mlx_lm.server --model mlx-community/Qwen3-VL-235B-A22B-Thinking-3bit

Official weights and documentation: huggingface.co/Qwen ↗

$ how the score was worked out · Qwen (Alibaba) vendor-reporte… 37%, Qwen/Qwen3.5-397B-A17B model… 28%, Qwen3-VL-235B-A22B-Thinking m… 26% and 1 more

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

source	results	share of the score	licence
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-reported	9	▇▇▇▇▇▇▇▇▇▇▇ 37%	Factual vendor-reported score (not licensed content)
Qwen/Qwen3.5-397B-A17B model card · lab-reported	25	▇▇▇▇▇▇▇▇ 28%	apache-2.0
Qwen3-VL-235B-A22B-Thinking model card · lab-reported	14	▇▇▇▇▇▇▇▇ 26%	Apache-2.0
LMArena leaderboard dataset	2	▇▇▇ 9%	CC BY 4.0
prior · pull toward the middle		▇ 0%
weighted average		= 64.1

$ improve this page

Update model information Add a place to run it Add a test result Propose a benchmark Correct a benchmark score Edit 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 2026-10-11 02:20 UTC · sources and licences · json