NeuroGenesysWhat LLMs can I run on a Mac with 48 GB?
Last updated • By Will Schulz
On a Mac with 48 GB of unified memory you can run 18 of the 24 open-weight models we track with an 8K context. The largest is Qwen2.5 72B at 2-bit (Q2_K), which needs 33.4 GB, and 8 run unquantized at 16-bit.
Models that fit a Mac with 48 GB of unified memory
| Model | Best fit | Memory | Longest context | |
|---|---|---|---|---|
| Qwen2.5 72B | 2-bit (Q2_K) | 33.4 GB | 16K | calculator |
| Llama 3.1 70B | 2-bit (Q2_K) | 32.5 GB | 16K | calculator |
| Llama 3.3 70B | 2-bit (Q2_K) | 32.5 GB | 16K | calculator |
| Mixtral 8x7B | Q5_K_M | 36.3 GB | 16K | calculator |
| Qwen3 32B | Q6_K | 30.9 GB | 32K | calculator |
| Qwen2.5 32B | Q6_K | 30.6 GB | 32K | calculator |
| Qwen3 30B-A3B | INT8 / Q8_0 | 35.2 GB | 32K | calculator |
| Gemma 3 27B | INT8 / Q8_0 | 32.1 GB | 64K | calculator |
| Gemma 2 27B | INT8 / Q8_0 | 33.1 GB | 32K | calculator |
| gpt-oss-20b | INT8 / Q8_0 | 24.2 GB | 128K | calculator |
| Qwen2.5 14B | BF16 / FP16 | 32.9 GB | 32K | calculator |
| Phi-4 (14B) | BF16 / FP16 | 33.0 GB | 32K | calculator |
| Gemma 3 12B | BF16 / FP16 | 27.0 GB | 128K | calculator |
| Gemma 2 9B | BF16 / FP16 | 22.1 GB | 64K | calculator |
| Qwen3 8B | BF16 / FP16 | 19.1 GB | 128K | calculator |
| Llama 3.1 8B | BF16 / FP16 | 18.6 GB | 128K | calculator |
| Qwen2.5 7B | BF16 / FP16 | 17.2 GB | 128K | calculator |
| Mistral 7B v0.3 | BF16 / FP16 | 17.0 GB | 128K | calculator |
Memory is weights + FP16 KV cache + runtime overhead for one request, as in the VRAM calculator, assuming macOS letting the GPU use 75% of unified memory (about 36.0 GB) by default. Longest context is the largest of 4K–128K that fits at that quantization; it is only usable up to the model's own context window.
Models too large for a Mac with 48 GB of unified memory
Even at 2-bit (Q2_K) these need more memory: Llama 3.1 405B, Llama 4 Scout, Llama 4 Maverick, Qwen3 235B-A22B, gpt-oss-120b, DeepSeek-V3 / R1.
Questions
What is the largest LLM that runs on a Mac with 48 GB of unified memory?
Of the models we track, Qwen2.5 72B (72.7B parameters) at 2-bit (Q2_K), which needs 33.4 GB with an 8K context. Larger models need more memory.
Can a Mac with 48 GB of unified memory run a model at full 16-bit precision?
Yes, 8 models: Qwen2.5 14B, Phi-4 (14B), Gemma 3 12B, Gemma 2 9B, Qwen3 8B, Llama 3.1 8B, Qwen2.5 7B, Mistral 7B v0.3. Larger models need quantizing, usually to Q4_K_M (about 4.9 bits per weight), which loses little quality.
How much memory does a 48 GB Mac leave for context?
The KV cache grows with every token of context, so the longest context that fits depends on the model and quantization; the table lists it for each model, trying 4K, 8K, 16K, 32K, 64K and 128K tokens. These figures assume macOS letting the GPU use 75% of unified memory (about 36.0 GB) by default.
