The Grux AI app's fit rule rates every model in its catalogue for a 48 GB Mac below, by the memory the model needs while it runs against about three quarters of the Mac's unified memory. The model the app starts a 48 GB Mac on is marked, with its Ollama pull line. The summary below the table is plain text to copy, and this address is the 48 GB result of the Mac picker, so it can be linked or shared. No speed is stated: speed depends on the chip, and only one Mac has been measured. A rating here says whether a model fits in memory, and nothing about how fast it runs or how well it answers.
48 GB Mac: about 36 GB is available to a model.
The app would start this Mac on Grux Local 9B (great fit): ollama pull gruxai.com/grux/grux-local:9b
| Model | Memory it needs | Fit |
|---|---|---|
| Grux Local 9BOllama 0.40.8 or newer | 7.25 GB | Great fit |
| Grux Local 4BOllama 0.40.8 or newer | 4.44 GB | Great fit |
| Grux Local 2BOllama 0.40.8 or newer | 2.65 GB | Great fit |
| Gemma 4 12BOllama 0.30.9 or newer | 9.4 GB | Great fit |
| Qwen 3.8 27BOllama 0.32.12 or newer | 17.4 GB | Great fit |
| Qwen 3.5 4BOllama 0.30.0 or newer | 4.4 GB | Great fit |
| Qwen 3.5 9BOllama 0.30.0 or newer | 6.2 GB | Great fit |
| Gemma 4 26BOllama 0.30.9 or newer | 19.2 GB | Good fit |
| Qwen 3.5 2BOllama 0.30.0 or newer | 2.6 GB | Great fit |
| GPT-OSS 20BOllama 0.30.0 or newer | 20 GB | Good fit |
| Mistral Small 3.2 24B | 22 GB | Good fit |
| Llama 3.2 3B | 3 GB | Great fit |
| Qwen 3 VL 4BOllama 0.30.0 or newer | 6 GB | Great fit |
| Qwen 3 Coder 30B | 28 GB | Tight |
Too big for this Mac: Qwen 3.5 35B (35 GB), Llama 3.3 70B (60 GB), GPT-OSS 120B (88 GB).
Select the box and copy it. It is the result above as plain text, and every rating in it is the Mac picker's.
Which local LLMs a 48 GB Mac can run, by the Grux AI app's fit rule (about 36 GB available to a model)
Start with: Grux Local 9B (ollama pull gruxai.com/grux/grux-local:9b)
Great fit: Grux Local 9B, Grux Local 4B, Grux Local 2B, Gemma 4 12B, Qwen 3.8 27B, Qwen 3.5 4B, Qwen 3.5 9B, Qwen 3.5 2B, Llama 3.2 3B, Qwen 3 VL 4B
Good fit: Gemma 4 26B, GPT-OSS 20B, Mistral Small 3.2 24B
Tight: Qwen 3 Coder 30B
Too big: Qwen 3.5 35B, Llama 3.3 70B, GPT-OSS 120B
Rated by memory only, not speed.
https://gruxai.com/answers/local-llm-for-a-48gb-mac/
From this size to the next one in the picker. The memory step chooser compares every larger size.
| Mac memory | The app starts it on | What changes from 48 GB |
|---|---|---|
| 64 GB16 GB more | Grux Local 9Bsame as on a 48 GB Mac | Gemma 4 26B: Good fit to Great fit; GPT-OSS 20B: Good fit to Great fit; Mistral Small 3.2 24B: Good fit to Great fit; Qwen 3 Coder 30B: Tight to Good fit; Qwen 3.5 35B: Too big to Good fit |
8 GB, 16 GB, 18 GB, 24 GB, 32 GB, 36 GB, 64 GB, 96 GB, 128 GB
| Claim | Source |
|---|---|
| The fit rule and every model's serving memory | Grux Cookbook.swift at the pinned commit |
| A Mac's model budget is a share of its unified memory, which Metal reports for each GPU | Apple, recommendedMaxWorkingSetSize |