Outmake Models
Models

Gemma 3 4B

Google · February 2025

Get Hugging Face

A 3.9B chat model from Google that reads pictures as well as text.

Rating
3.1
Downloads
423K
Likes
752
Parameters
3.9B
Context
128K
Licence
gemma

Runs great on M4 · 16 GB

The Q6_K_L size from bartowski is the one to get: a 3.4 GB download that uses 5.1 GB of memory, and writes about 16 words a second.

Ratings

3.1 out of 5, across people's blind votes on its answers.

Text
1303
Coding
1273

LM Arena ratings of October 28, 2026 (CC BY 4.0): people vote between two answers without knowing which model wrote which. The bar ends at 1488, the best any open model has. The Arena tests the full model; a small size on a Mac answers a little worse.

Gemma 3 4B runs well from a Mac with an M1 and 8 GB of memory. 59 of 59 Mac configurations run it well.

Compare with

The nearest models in size.

Sizes

From unsloth/gemma-3-4b-it-qat-GGUF on Hugging Face. A smaller size is a smaller download and a little less accurate.

SizeDownloadMemory at 8KGet
UD-IQ1_S1.2 GB1.4 GBestimatedOpen
UD-IQ1_M1.2 GB1.5 GBestimatedOpen
UD-IQ2_XXS1.3 GB1.6 GBestimatedOpen
UD-IQ2_M1.6 GB1.8 GBestimatedOpen
UD-IQ3_XXS1.7 GB2.0 GBestimatedOpen
Q2_K1.7 GB2.0 GBestimatedOpen
Q2_K_L1.7 GB2.0 GBestimatedOpen
UD-Q2_K_XL1.8 GB2.0 GBestimatedOpen
Q3_K_S1.9 GB2.2 GBestimatedOpen
Q3_K_M2.1 GB2.4 GBestimatedOpen
UD-Q3_K_XL2.2 GB2.4 GBestimatedOpen
IQ4_XS2.3 GB2.5 GBestimatedOpen
IQ4_NL2.4 GB2.6 GBestimatedOpen
Q4_02.4 GB2.6 GBestimatedOpen
Q4_K_S2.4 GB2.6 GBestimatedOpen
Q4_K_M2.5 GB2.8 GBestimatedOpen
UD-Q4_K_XL2.5 GB2.8 GBestimatedOpen
Q4_12.6 GB2.8 GBestimatedOpen
Q5_K_S2.8 GB3.0 GBestimatedOpen
Q5_K_M2.8 GB3.1 GBestimatedOpen
UD-Q5_K_XL2.8 GB3.1 GBestimatedOpen
Q6_K3.2 GB3.5 GBestimatedOpen

Run It Anywhere

Gemma 3 4B is an open model, so any app that runs GGUF files can run it. These take the Q6_K_L size from bartowski straight from Hugging Face.

  • LM StudioMac, Windows, LinuxOpen
  • JanMac, Windows, LinuxOpen
  • MstyMac, Windows, LinuxOpen
  • OllamaMac, Windows, Linuxollama run hf.co/bartowski/google_gemma-3-4b-it-qat-GGUF:Q6_K_L
  • llama.cppMac, Windows, Linuxllama-server -hf bartowski/google_gemma-3-4b-it-qat-GGUF:Q6_K_L
  • Docker Model RunnerMac, Windows, Linuxdocker model run hf.co/bartowski/google_gemma-3-4b-it-qat-GGUF:Q6_K_L
On a PC Graphics Card
Video memoryLargest size that fitsDownloadMemory at 8K
6 GBQ6_K3.2 GB3.5 GB
8 GBQ6_K3.2 GB3.5 GB
10 GBQ6_K3.2 GB3.5 GB
11 GBQ6_K3.2 GB3.5 GB
12 GBQ6_K3.2 GB3.5 GB
16 GBQ6_K3.2 GB3.5 GB
20 GBQ6_K3.2 GB3.5 GB
24 GBQ6_K3.2 GB3.5 GB
32 GBQ6_K3.2 GB3.5 GB
48 GBQ6_K3.2 GB3.5 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
MacSizeFitMemoryWrites
M6 · 16 GBQ6_K_LRuns great5.1 GB24
M6 · 24 GBQ6_K_LRuns great5.1 GB24
M6 · 32 GBQ6_K_LRuns great5.1 GB24
M5 · 16 GBQ6_K_LRuns great5.1 GB22
M5 · 24 GBQ6_K_LRuns great5.1 GB22
M5 · 32 GBQ6_K_LRuns great5.1 GB22
M5 Pro · 24 GBQ6_K_LRuns great5.1 GB42
M5 Pro · 48 GBQ6_K_LRuns great5.1 GB42
M5 Pro · 64 GBQ6_K_LRuns great5.1 GB42
M5 Max · 36 GBQ6_K_LRuns great5.1 GB60
M5 Max · 48 GBQ6_K_LRuns great5.1 GB77
M5 Max · 64 GBQ6_K_LRuns great5.1 GB77
M5 Max · 128 GBQ6_K_LRuns great5.1 GB77
M5 Ultra · 96 GBQ6_K_LRuns great5.1 GB133
M5 Ultra · 256 GBQ6_K_LRuns great5.1 GB133
M5 Ultra · 512 GBQ6_K_LRuns great5.1 GB133
M4 · 16 GBQ6_K_LRuns great5.1 GB16
M4 · 24 GBQ6_K_LRuns great5.1 GB16
M4 · 32 GBQ6_K_LRuns great5.1 GB16
M4 Pro · 24 GBQ6_K_LRuns great5.1 GB31
M4 Pro · 48 GBQ6_K_LRuns great5.1 GB31
M4 Pro · 64 GBQ6_K_LRuns great5.1 GB31
M4 Max · 36 GBQ6_K_LRuns great5.1 GB41
M4 Max · 48 GBQ6_K_LRuns great5.1 GB49
M4 Max · 64 GBQ6_K_LRuns great5.1 GB49
M4 Max · 96 GBQ6_K_LRuns great5.1 GB41
M4 Max · 128 GBQ6_K_LRuns great5.1 GB49
M3 · 8 GBQ6_K_LRuns great3.7 GB13
M3 · 16 GBQ6_K_LRuns great5.1 GB13
M3 · 24 GBQ6_K_LRuns great5.1 GB13
M3 Pro · 18 GBQ6_K_LRuns great5.1 GB19
M3 Pro · 36 GBQ6_K_LRuns great5.1 GB19
M3 Max · 36 GBQ6_K_LRuns great5.1 GB34
M3 Max · 48 GBQ6_K_LRuns great5.1 GB42
M3 Max · 64 GBQ6_K_LRuns great5.1 GB42
M3 Max · 96 GBQ6_K_LRuns great5.1 GB34
M3 Max · 128 GBQ6_K_LRuns great5.1 GB42
M3 Ultra · 96 GBQ6_K_LRuns great5.1 GB66
M3 Ultra · 256 GBQ6_K_LRuns great5.1 GB66
M3 Ultra · 512 GBQ6_K_LRuns great5.1 GB66
M2 · 8 GBQ6_K_LRuns great3.7 GB13
M2 · 16 GBQ6_K_LRuns great5.1 GB13
M2 · 24 GBQ6_K_LRuns great5.1 GB13
M2 Pro · 16 GBQ6_K_LRuns great5.1 GB25
M2 Pro · 32 GBQ6_K_LRuns great5.1 GB25
M2 Max · 32 GBQ6_K_LRuns great5.1 GB42
M2 Max · 64 GBQ6_K_LRuns great5.1 GB42
M2 Max · 96 GBQ6_K_LRuns great5.1 GB42
M2 Ultra · 64 GBQ6_K_LRuns great5.1 GB65
M2 Ultra · 128 GBQ6_K_LRuns great5.1 GB65
M2 Ultra · 192 GBQ6_K_LRuns great5.1 GB65
M1 · 8 GBQ6_K_LRuns great3.7 GB9
M1 · 16 GBQ6_K_LRuns great5.1 GB9
M1 Pro · 16 GBQ6_K_LRuns great5.1 GB22
M1 Pro · 32 GBQ6_K_LRuns great5.1 GB22
M1 Max · 32 GBQ6_K_LRuns great5.1 GB35
M1 Max · 64 GBQ6_K_LRuns great5.1 GB35
M1 Ultra · 64 GBQ6_K_LRuns great5.1 GB50
M1 Ultra · 128 GBQ6_K_LRuns great5.1 GB50

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

Packaged by

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

Information

Maker
Google
Released
February 2025
Parameters
3.9B
Longest chat
128K tokens
Reads pictures
Yes
Licence
gemma
Downloads this month
422,789
Likes
752
Sizes
22, 1.2 to 3.2 GB
Kind
Maker's release

Badge for the model card

Shows on Gemma 3 4B's Hugging Face card which Macs run it well, linked to this page. It updates every morning.

Runs on Mac badge

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.

Your Computer

Every model is judged for this Mac. Both are under About This Mac, in the Apple menu.

Every Mac · Every graphics card

Civitai

Open in OutmakeHugging Face

New to Outmake? Download the free beta