Models

Mistral Small 4 119B 2603

Mistral AI · January 2026

Get

Hugging Face

A 119B, 6.6B active chat model from Mistral AI that reads pictures as well as text.

Downloads: 46K
Likes: 114
Parameters: 119B, 6.6B active
Context: 1M
Licence: apache-2.0

Too big for this Mac

Its smallest size needs 41 GB of memory, more than M4 · 16 GB can give it.

Ratings

ScienceGPQA Diamond: 71.2%

GPQA Diamond: questions PhD scientists write for each other. Scores are the ones filed on the model's Hugging Face page or the board's own table, read 2026-11-05; some come from the maker.

Mistral Small 4 119B 2603 runs well from a Mac with an M3 Max and 48 GB of memory. 29 of 59 Mac configurations run it well.

Compare with

The nearest models in size.

Sizes

From unsloth/Mistral-Small-4-119B-2603-GGUF on Hugging Face. A smaller size is a smaller download and a little less accurate.

Size	Download	Memory at 8K	Get
UD-IQ1_M	32 GB	33 GBestimated	Open
UD-IQ2_XXS	35 GB	35 GBestimated	Open
UD-IQ2_M	38 GB	38 GBestimated	Open
UD-Q2_K_XL	40 GB	40 GBestimated	Open
UD-IQ3_XXS	43 GB	43 GBestimated	Open
UD-IQ3_S	44 GB	45 GBestimated	Open
UD-Q3_K_S	50 GB	50 GBestimated	Open
UD-Q3_K_M	54 GB	55 GBestimated	Open
UD-Q3_K_XL	55 GB	55 GBestimated	Open
UD-IQ4_XS	58 GB	58 GBestimated	Open
UD-IQ4_NL	59 GB	59 GBestimated	Open
UD-Q4_K_S	69 GB	70 GBestimated	Open
MXFP4_MOE	72 GB	72 GBestimated	Open
UD-Q4_K_M	74 GB	74 GBestimated	Open
UD-Q4_K_XL	74 GB	74 GBestimated	Open
UD-Q5_K_S	84 GB	84 GBestimated	Open
UD-Q5_K_M	89 GB	89 GBestimated	Open
UD-Q5_K_XL	89 GB	90 GBestimated	Open
UD-Q6_K	99 GB	100 GBestimated	Open
UD-Q6_K_XL	108 GB	108 GBestimated	Open
Q8_0	126 GB	127 GBestimated	Open
UD-Q8_K_XL	133 GB	133 GBestimated	Open
BF16	238 GB	238 GBestimated	Open

Run It Anywhere

Mistral Small 4 119B 2603 is an open model, so any app that runs GGUF files can run it. These take the UD-IQ1_M size from unsloth straight from Hugging Face.

LM StudioMac, Windows, LinuxOpen
JanMac, Windows, LinuxOpen
MstyMac, Windows, LinuxOpen
OllamaMac, Windows, Linuxollama run hf.co/unsloth/Mistral-Small-4-119B-2603-GGUF:UD-IQ1_M
llama.cppMac, Windows, Linuxllama-server -hf unsloth/Mistral-Small-4-119B-2603-GGUF:UD-IQ1_M
Docker Model RunnerMac, Windows, Linuxdocker model run hf.co/unsloth/Mistral-Small-4-119B-2603-GGUF:UD-IQ1_M

On a PC Graphics Card

Video memory	Largest size that fits	Download	Memory at 8K
6 GB	None worth running		needs 38 GB
8 GB	None worth running		needs 38 GB
10 GB	None worth running		needs 38 GB
11 GB	None worth running		needs 38 GB
12 GB	None worth running		needs 38 GB
16 GB	None worth running		needs 38 GB
20 GB	None worth running		needs 38 GB
24 GB	None worth running		needs 38 GB
32 GB	None worth running		needs 38 GB
48 GB	UD-IQ3_S	44 GB	45 GB

The largest size, of those that keep three bits a weight or more, whose memory at an 8K chat is under 92% of the card's; the rest is the desktop's. Estimated from the files, the same way as for a Mac; speed on a PC is not estimated yet. A size that doesn't fit the card still runs, slower, with part of it in ordinary memory.

On every Mac

Mac	Size	Fit	Memory	Writes
M6 · 16 GB	IQ1_S	Too big	41 GB
M6 · 24 GB	IQ1_S	Too big	41 GB
M6 · 32 GB	IQ1_S	Too big	41 GB
M5 · 16 GB	IQ1_S	Too big	41 GB
M5 · 24 GB	IQ1_S	Too big	41 GB
M5 · 32 GB	IQ1_S	Too big	41 GB
M5 Pro · 24 GB	IQ1_S	Too big	41 GB
M5 Pro · 48 GB	IQ1_M	Runs well	31 GB	45
M5 Pro · 64 GB	IQ1_M	Runs great	31 GB	45
M5 Max · 36 GB	IQ1_S	Tight fit	28 GB	66
M5 Max · 48 GB	IQ1_M	Runs well	31 GB	78
M5 Max · 64 GB	IQ1_M	Runs great	31 GB	78
M5 Max · 128 GB	IQ4_NL	Runs great	68 GB	47
M5 Ultra · 96 GB	IQ4_NL	Runs well	68 GB	81
M5 Ultra · 256 GB	Q8_0	Runs great	141 GB	62
M5 Ultra · 512 GB	Q8_0	Runs great	141 GB	62
M4 · 16 GB	IQ1_S	Too big	41 GB
M4 · 24 GB	IQ1_S	Too big	41 GB
M4 · 32 GB	IQ1_S	Too big	41 GB
M4 Pro · 24 GB	IQ1_S	Too big	41 GB
M4 Pro · 48 GB	IQ1_M	Runs well	31 GB	30
M4 Pro · 64 GB	IQ1_M	Runs great	31 GB	30
M4 Max · 36 GB	IQ1_S	Tight fit	28 GB	39
M4 Max · 48 GB	IQ1_M	Runs well	31 GB	44
M4 Max · 64 GB	IQ1_M	Runs great	31 GB	44
M4 Max · 96 GB	IQ4_NL	Runs well	68 GB	26
M4 Max · 128 GB	IQ4_NL	Runs great	68 GB	31
M3 · 8 GB	IQ1_S	Too big	41 GB
M3 · 16 GB	IQ1_S	Too big	41 GB
M3 · 24 GB	IQ1_S	Too big	41 GB
M3 Pro · 18 GB	IQ1_S	Too big	41 GB
M3 Pro · 36 GB	IQ1_S	Tight fit	28 GB	22
M3 Max · 36 GB	IQ1_S	Tight fit	28 GB	35
M3 Max · 48 GB	IQ1_M	Runs well	31 GB	41
M3 Max · 64 GB	IQ1_M	Runs great	31 GB	41
M3 Max · 96 GB	IQ4_NL	Runs well	68 GB	22
M3 Max · 128 GB	IQ4_NL	Runs great	68 GB	27
M3 Ultra · 96 GB	IQ4_NL	Runs well	68 GB	42
M3 Ultra · 256 GB	Q8_0	Runs great	141 GB	34
M3 Ultra · 512 GB	Q8_0	Runs great	141 GB	34
M2 · 8 GB	IQ1_S	Too big	41 GB
M2 · 16 GB	IQ1_S	Too big	41 GB
M2 · 24 GB	IQ1_S	Too big	41 GB
M2 Pro · 16 GB	IQ1_S	Too big	41 GB
M2 Pro · 32 GB	IQ1_S	Too big	41 GB
M2 Max · 32 GB	IQ1_S	Too big	41 GB
M2 Max · 64 GB	IQ1_M	Runs great	31 GB	40
M2 Max · 96 GB	IQ4_NL	Runs well	68 GB	27
M2 Ultra · 64 GB	IQ1_M	Runs great	31 GB	57
M2 Ultra · 128 GB	IQ4_NL	Runs great	68 GB	42
M2 Ultra · 192 GB	Q8_0	Runs great	130 GB	34
M1 · 8 GB	IQ1_S	Too big	41 GB
M1 · 16 GB	IQ1_S	Too big	41 GB
M1 Pro · 16 GB	IQ1_S	Too big	41 GB
M1 Pro · 32 GB	IQ1_S	Too big	41 GB
M1 Max · 32 GB	IQ1_S	Too big	41 GB
M1 Max · 64 GB	IQ1_M	Runs great	31 GB	32
M1 Ultra · 64 GB	IQ1_M	Runs great	31 GB	41
M1 Ultra · 128 GB	IQ4_NL	Runs great	68 GB	32

Writes is words a second, estimated from the Mac's memory speed and the model's size; how, and how wrong.

Packaged by

4 repositories on Hugging Face carry this model for llama.cpp. Outmake reads them all as one model.

Information

Maker: Mistral AI
Released: January 2026
Parameters: 119B, 6.6B active
Longest chat: 1M tokens
Reads pictures: Yes
Licence: apache-2.0
Downloads this month: 45,716
Likes: 114
Sizes: 23, 32 to 238 GB
Kind: Maker's release
Source: mistralai/Mistral-Small-4-119B-2603

Badge for the model card

Shows on Mistral Small 4 119B 2603's Hugging Face card which Macs run it well, linked to this page. It updates every morning.

Files, downloads and likes read from Hugging Face on November 5, 2026 and refreshed every morning. Memory and speed are estimates until someone running Outmake measures this model on that Mac.