whatmodel.sh / Pixtral 12B (2409)

Pixtral 12B (2409)

+ contribute

Pixtral 12B (2409) is an open-weights language model from Mistral, released September 2024. Its published size is 12.4B parameters. Licence: apache-2.0.

Catalogue identity notes

Comparison-only. Arena entry 'pixtral-12b-2409' (vision). LMArena's leaderboard table (arena-leaderboard space @6f2eeeb) labels 'pixtral-12b-2409' as 'Pixtral-12B-2409' (Mistral, Apache 2.0), linking https://mistral.ai/news/pixtral-12b/. Maker repo mistralai/Pixtral-12B-2409 @a5ca2b7: licence tag apache-2.0. Parameters: card: '12B parameter Multimodal Decoder + 400M parameter Vision Encoder'. Released 2024-09: month the maker's Hugging Face repo was created (createdAt 2024-09-11).

48.8±1.8#365 of 719 LLMsfrom 11 benchmark results across 4 sources

$ how it compares

The LLMs ranked either side of Pixtral 12B (2409), out of 828 with a score, then more from Mistral.

#	model	score	where	RAM	released
434	Qwen3-VL 2B Thinking Alibaba	49±1	local	~2.7GB	2025-10
435	speechless-codellama-34b-v2.0 uukuguy	49±1	local	~23GB	2023-10
436	Pixtral 12B (2409) Mistral	49±2	local	~9.6GB	2024-09
437	Nova Lite Amazon	49±2	api $0.10/M		2024-12
438	MegaBeam-Mistral-7B-512k AWS (aws-prototyping)	49±4	local	~6.1GB	2024-07
	more from Mistral
79	Mistral Large 4 (public preview) Mistral	71±2	api $2.06/M		2026-10
174	Mistral Medium 3.5 (128B) Mistral	65±1	api $3.00/M +local	~42GB	2026-04
180	Mistral Medium 3.1 (2508) Mistral	64±2	comparison only		2025-08

see all LLMs · find the ones that fit your machine

$ what it's good at

coding▁▁▁▃▆██▇▆▄▃▁ #276 of 458 * · 21% of its score · 2 tests

Beats Qwen3-VL 32B Instruct, Solar Open 100B (102B-A12B), GLM-4.7-Flash here, though they rank higher overall. Trails DeepSeek LLM 67B Chat, Yi 1.5 9B Chat, Mistral NeMo Instruct (2407) here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
VisEditBench strict pass rate (1,395 visualization-code editing tasks)	35.3% ✎	▄··█·▄··▄█·▄ beats 2/7	hard · avg 34%	▇▇▇▇▇ 14%	at its level
HumanEval lab	72.0% ✎	▁▁▁▁▁▃▅▅█▃▂▁ beats 128/190	easy · avg 72%	▇▇▇ 7.5%	sharp · at its level · noisy

math▁▂▃▅▆▅▄█▄▁▁▁ #283 of 398 * · 15% of its score · 2 tests

Beats Llama 3.1 Nemotron 70B Instruct, GPT-4 (0613), Athene-70B here, though they rank higher overall. Trails DeepSeek R1 Distill Qwen 1.5B, Phi-2 (2.7B), Granite 8B Code Base (4K) here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
MATH (subset unspecified) lab	48.1% ✎	▂▄▄▃█▇▅▆▄▃▄▁ beats 24/80	medium · avg 55%	▇▇▇▇ 8.9%	at its level
MathVista (testmini) lab	58.3% ✎	▂····▂···▂█▆ beats 6/9	medium · avg 53%	▇▇▇ 6.4%	at its level

vision & documents▁▁▃▃▁▃▄▂▆▇█▄ #123 of 147 · 65% of its score · 8 tests

Beats Gemma 3 12B, Llama 3.2 90B Vision Instruct, Nova Pro here, though they rank higher overall. Trails Qwen3.5 2B here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
Arena vision (style controlled)	1025.2 ✎	▁▁▂▂▃▂▄▅▅▆██ beats 17/145	avg 1068	▇▇▇▇▇▇▇▇ 19%	blunt
VisEditBench strict pass rate (1,395 visualization-code editing tasks)	35.3% ✎	▄··█·▄··▄█·▄ beats 2/7	hard · avg 34%	▇▇▇▇▇ 14%	at its level
MM-MT-Bench (Mistral, GPT-4o judge) lab	6.0500 MM-MT-Bench score (1-10) ✎	▅··▃·█·▃▃··▅ beats 6/9	avg 5.740 MM-MT-Bench score (1-10)	▇▇▇▇ 9.5%	sharp · noisy
MMMU (val) lab	52.0% ✎	▂·▂······█▄▄ beats 4/9	medium · avg 52%	▇▇▇ 6.7%	at its level
MathVista (testmini) lab	58.3% ✎	▂····▂···▂█▆ beats 6/9	medium · avg 53%	▇▇▇ 6.4%	at its level
VQAv2 (val) lab	78.6% ✎	▃···█▃··▅▃▃▃ beats 8/9	easy · avg 74%	▇▇ 5.6%	blunt · at its level
Ai2 Molmo human preference Elo (Sept 2024, 27 VLMs)	1016.0 ✎	▄▄·▄·▄▄▄·▄·█ beats 3/8	avg 1037	▇ 3.1%	blunt
ChartQA lab	81.8% ✎	▄·▄▄▄··▄▄▄▄█ beats 7/9	easy · avg 77%	▇ 0.5%	blunt · noisy

other tests1 not in a task · 18% of its score

test	result	vs the others in its run	difficulty	counts	why
MMLU lab	69.2% ✎	▁▁▁▁▁▂▃▅█▃▆▃ beats 110/183	easy · avg 76%	▇▇▇▇▇▇▇ 18%	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 Pixtral 12B (2409) · needs ~7.7 GB · from huggingface.co/EnlistedGhost
chat in the terminal
llama cli -hf EnlistedGhost/Pixtral-12B-2409-GGUF:Q4_K_M
Or start a local server and open the address it prints (http://localhost:9931)
llama serve -hf EnlistedGhost/Pixtral-12B-2409-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 Pixtral 12B (2409) · needs ~7.7 GB · from huggingface.co/EnlistedGhost
open in LM Studio ↗
Or download it in the terminal
lms get https://huggingface.co/EnlistedGhost/Pixtral-12B-2409-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 Pixtral 12B (2409) · needs ~7.7 GB · from huggingface.co/EnlistedGhost
chat in the terminal
ollama run hf.co/EnlistedGhost/Pixtral-12B-2409-GGUF:Q4_K_M

Other ways to run it

Run it as published · needs ~25.37 GB, NVIDIA only · from huggingface.co/mistralai
pip install --upgrade vllm mistral_common && vllm serve mistralai/Pixtral-12B-2409 --tokenizer_mode mistral --limit_mm_per_prompt 'image=4'

Weights and documentation: huggingface.co/mistralai ↗

$ how the score was worked out · Pixtral 12B (Mistral AI, arXi… 63%, LMArena leaderboard dataset 19%, VisEditBench: Can Vision-Lang… 14% 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
Pixtral 12B (Mistral AI, arXiv 2410.07073v2) · lab-reported	8	▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 63%	CC BY 4.0
LMArena leaderboard dataset	1	▇▇▇▇▇▇ 19%	CC BY 4.0
VisEditBench: Can Vision-Language Models Edit Visualization Code from Multimodal Feedback? (arXiv:2608.10408v1)	1	▇▇▇▇ 14%	CC BY 4.0
Ai2 Molmo-72B-0924 model card: Human Preference Elo Rating column	1	▇ 3%	Apache-2.0
prior · pull toward the middle		▇ 1%
weighted average		= 48.8

$ 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