whatmodel.sh / Qwen2-VL 72B Instruct

Qwen2-VL 72B Instruct

+ contribute

Qwen2-VL 72B Instruct is an open-weights language model from Alibaba, released September 2024. Its published size is 73.4B parameters. Licence: tongyi-qianwen.

Catalogue identity notes

Comparison-only. Arena entry 'qwen2-vl-72b' (vision). LMArena's leaderboard table (arena-leaderboard space @6f2eeeb) labels 'qwen2-vl-72b' as 'Qwen2-VL-72b-Instruct' (Alibaba, Qwen), linking https://qwenlm.github.io/zh/blog/qwen2-vl/. Maker repo Qwen/Qwen2-VL-72B-Instruct @2ac26c9: licence tag other (tongyi-qianwen). 73,405,560,320 parameters from the Hugging Face safetensors metadata (dense, so active = total). Released 2024-09: month the maker's Hugging Face repo was created (createdAt 2024-09-17). Arena key 'qwen2-vl-72b'; LMArena's table labels it 'Qwen2-VL-72b-Instruct'.

53.2±1.0#294 of 719 LLMsfrom 29 benchmark results across 6 sources

$ how it compares

The LLMs ranked either side of Qwen2-VL 72B Instruct, out of 828 with a score, then more from Alibaba.

#	model	score	where	RAM	released
349	GLM-4 9B Chat 1M Zhipu AI (THUDM)	53±5	local	~8.1GB	2024-06
350	Mistral Small 3 24B (2501) Mistral	53±1	local	~16GB	2025-01
351	Qwen2-VL 72B Instruct Alibaba	53±1	local	~34GB	2024-09
352	Gemini 1.5 Flash Experimental (0827) Google	53±1	comparison only		2024-08
353	GPT-4o mini (2024-07-18) OpenAI	53±0	api $0.26/M		2024-07
	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

math▁▂▃▅▆▅▄█▄▁▁▁ #246 of 398 · 17% of its score · 2 tests

Beats GPT-4o (2024-11-20), GPT-4 Turbo (2024-04-09), Molmo 2 8B here, though they rank higher overall. Trails DeepSeek R1 Distill Qwen 1.5B, Tulu 2 DPO 70B, OpenThinker 7B here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
MathVista (testmini) lab +1	70.5% ✎	▃··▃▅·▃█▅▅▅█ beats 5/16	medium · avg 69%	▇▇▇▇▇ 12%	at its level · several runs
MATH-Vision lab	25.9% ✎	▁▁·▃··▄▁▃▇▇█ beats 2/24	hard · avg 27%	▇▇ 5.1%	at its level · small run

vision & documents▁▁▃▂▁▃▄▂▆▇█▄ #116 of 147 · 87% of its score · 18 tests

Beats DeepSeek V4 Flash Vision (Exp), GPT-4 Turbo (2024-04-09) here, though they rank higher overall. Trails Qwen3.5 2B, Qwen3-VL 4B Instruct, Gemma 4 E4B here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
MMMU (val) lab +1	64.5% ✎	▂···▆▂▂▃▄▄▄█ beats 6/16	medium · avg 61%	▇▇▇▇▇▇ 16%	at its level · several runs
MathVista (testmini) lab +1	70.5% ✎	▃··▃▅·▃█▅▅▅█ beats 5/16	medium · avg 69%	▇▇▇▇▇ 12%	at its level · several runs
MMStar lab +1	68.3% ✎	▂··▄▂▂·▄█▂▄▆ beats 6/16	medium · avg 65%	▇▇▇▇▇ 11%	at its level · several runs
MMMU Pro lab +1	46.2% ✎	▃·█·▅▃▃▃▅▅▃█ beats 6/16	medium · avg 46%	▇▇▇▇ 11%	at its level · several runs
Arena vision (style controlled)	1084.8 ✎	▁▁▂▂▂▃▄▅▅▆██ beats 27/145	avg 1068	▇▇ 5.8%	blunt
MATH-Vision lab	25.9% ✎	▁▁·▃··▄▁▃▇▇█ beats 2/24	hard · avg 27%	▇▇ 5.1%	at its level · small run
CapArena-Auto detailed image captioning (Score_Avg, GPT-4o judge)	-9.0000 ✎	▃▃▃·▃█·▅▃▃▃▃ beats 5/12	avg -0.671	▇▇ 3.9%	typical
DocVQA lab +1	96.5% ✎	▃··▅▃▅▃█▃█▅▅ beats 14/17	very easy · avg 94%	▇ 3.6%	blunt · maxed out · several runs

10 more tests

test	result	vs the others in its run	difficulty	counts	why
HallusionBench lab	58.1% ✎	▃·▃·▅▃▅▅▃▅█▅ beats 8/16	medium · avg 55%	▇ 3.5%	blunt · at its level
ChartQA lab +1	88.3% ✎	▂··▂·▂▃·▅▃▆█ beats 8/17	easy · avg 85%	▇ 2.9%	blunt · several runs
MM-MT-Bench (Mistral, GPT-4o judge)	6.5900 MM-MT-Bench score (1-10) ✎	▅··▃·█·▃▃··▅ beats 7/9	avg 5.740 MM-MT-Bench score (1-10)	▇ 2.9%	sharp · noisy
InfographicVQA lab	84.5% ✎	▃·▃▅█▃▃▃▃▃▃▃ beats 10/13	easy · avg 80%	▇ 2.6%	small run
RealWorldQA lab	77.8% ✎	▅▃·▃▅▅·█·▃▅█ beats 13/16	easy · avg 70%	▇ 2.5%	blunt · at its level
VQAv2 (val)	81.6% ✎	▃···█▃··▅▃▃▃ beats 9/9	easy · avg 74%	▇ 1.5%	blunt
Ai2 Molmo human preference Elo (Sept 2024, 27 VLMs)	1037.0 ✎	▄▄·▄·▄▄▄·▄·█ beats 5/8	avg 1037	▇ 0.9%	blunt
OCRBench lab	87.7% ✎	▅█·▃▃█▃▅▅▃·▃ beats 10/16	easy · avg 84%	▇ 0.9%	blunt
CC-OCR lab	68.7% ✎	▁·▃··▂▂▃▃▂▆█ beats 2/26	easy · avg 72%	▇ 0.7%	blunt · at its level · small run
TextVQA lab	85.5% ✎	█······▄·▄▄▄ beats 5/5	easy · avg 80%	▇ 0.5%	blunt

other tests3 not in a task · 13% of its score

test	result	vs the others in its run	difficulty	counts	why
Video-MME (without subtitles) lab +1	71.2% ✎	▃··█··▅█▃▃▅▃ beats 6/13	medium · avg 66%	▇▇▇ 6.4%	blunt · at its level · several runs
Video-MME (with subtitles) lab +1	77.8% ✎	▁▁▁▁▁·▁▃▁▁▁█ beats 6/16	medium · avg 73%	▇▇ 4.5%	at its level · small run · several runs
MVBench lab	73.6% ✎	▂▂▂·▂█··▆▄·▂ beats 12/13	medium · avg 67%	▇ 2.0%	blunt · at its level

$ 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 Qwen2-VL 72B Instruct · needs ~48.82 GB · from huggingface.co/bartowski
chat in the terminal
llama cli -hf bartowski/Qwen2-VL-72B-Instruct-GGUF:Q4_K_M
Or start a local server and open the address it prints (http://localhost:9931)
llama serve -hf bartowski/Qwen2-VL-72B-Instruct-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 Qwen2-VL 72B Instruct · needs ~48.82 GB · from huggingface.co/bartowski
open in LM Studio ↗
Or download it in the terminal
lms get https://huggingface.co/bartowski/Qwen2-VL-72B-Instruct-GGUF@Q4_K_M -y
Then run lms ls to see its name, and lms chat followed by that name

Ollama

Install Ollama
download it (macOS / Windows) ↗
Linux
curl -fsSL https://ollama.com/install.sh | sh
Run Qwen2-VL 72B Instruct · needs ~31.21 GB · from huggingface.co/bartowski
chat in the terminal
ollama run hf.co/bartowski/Qwen2-VL-72B-Instruct-GGUF:Q2_K

MLX (Mac)

version

Install MLX (Mac) · Apple silicon Macs only, text only.
with uv
uv tool install mlx-lm
or
pip install mlx-lm
Run Qwen2-VL 72B Instruct · needs ~41.3 GB, Mac only · from huggingface.co/mlx-community
chat in the terminal
mlx_lm.chat --model mlx-community/Qwen2-VL-72B-Instruct-4bit
Or start an OpenAI-compatible server at http://localhost:8080/v1
mlx_lm.server --model mlx-community/Qwen2-VL-72B-Instruct-4bit

Official weights and documentation: huggingface.co/Qwen ↗

$ how the score was worked out · Qwen (Alibaba) vendor-reporte… 69%, Qwen2.5-VL-32B-Instruct model… 11%, Pixtral 12B (Mistral AI, arXi… 9% and 3 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	15	▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 69%	Factual vendor-reported score (not licensed content)
Qwen2.5-VL-32B-Instruct model card · lab-reported	6	▇▇▇ 11%	Apache-2.0
Pixtral 12B (Mistral AI, arXiv 2410.07073v2)	5	▇▇▇ 9%	CC BY 4.0
LMArena leaderboard dataset	1	▇▇ 6%	CC BY 4.0
CapArena: Benchmarking and Analyzing Detailed Image Captioning in the LLM Era (arXiv:2503.12329v1)	1	▇ 4%	CC-BY-4.0
Ai2 Molmo-72B-0924 model card: Human Preference Elo Rating column	1	▇ 1%	Apache-2.0
prior · pull toward the middle		▇ 0%
weighted average		= 53.2

$ 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