What LLM can a Mac with 16 GB unified memory run?
16 GB of unified memory is not 16 GB of VRAM. macOS lets the GPU wire about 67% by default, so this page budgets 10.7 GiB for weights, KV cache and runtime. The largest model in our list that fits is Mistral-Nemo-Instruct-2407 at Q4_K_M (8.3 GiB–8.9 GiB at 8,192 tokens). 8 of 22 models run, 3 are tight, 11 do not fit.
Model by model on 16 GB
| Model | Result | Quant to use | Estimated total | Max context |
|---|---|---|---|---|
| Phi-4-mini-instruct | Runs | Q8_0 | 4.9 GiB–5.5 GiB | 39,963 |
| Qwen3-4B | Runs | Q8_0 | 5.2 GiB–5.8 GiB | 34,257 |
| Mistral-7B-Instruct-v0.3 | Runs | Q8_0 | 8.3 GiB–8.9 GiB | 13,650 |
| Qwen2.5-7B-Instruct | Runs | Q8_0 | 8.1 GiB–8.7 GiB | 24,618 |
| DeepSeek-R1-Distill-Llama-8B | Runs | Q6_K | 7.3 GiB–7.8 GiB | 21,698 |
| Qwen3-8B | Runs | Q6_K | 7.5 GiB–8.1 GiB | 18,447 |
| GLM-4-9B-0414 | Runs | Q6_K | 8.1 GiB–8.7 GiB | 30,553 |
| Mistral-Nemo-Instruct-2407 | Runs | Q4_K_M | 8.3 GiB–8.9 GiB | 12,231 |
| phi-4 | Tight | Q4_K_M | 10.0 GiB–10.6 GiB | 11,858 |
| Qwen3-14B | Tight | Q4_K_M | 9.8 GiB–10.4 GiB | 14,143 |
| Qwen2.5-14B-Instruct | Tight | Q4_K_M | 10.0 GiB–10.6 GiB | 11,855 |
| Mistral-Small-24B-Instruct-2501 | No | Smallest file is 13.3 GiB | n/a | n/a |
| Qwen3-30B-A3B | No | Smallest file is 17.3 GiB | n/a | n/a |
| Qwen3-32B | No | Smallest file is 18.4 GiB | n/a | n/a |
| Qwen2.5-32B-Instruct | No | Smallest file is 18.5 GiB | n/a | n/a |
| DeepSeek-R1-Distill-Qwen-32B | No | Smallest file is 18.5 GiB | n/a | n/a |
| Mixtral-8x7B-Instruct-v0.1 | No | Smallest file is 26.5 GiB | n/a | n/a |
| Llama-3.1-Nemotron-70B-Instruct-HF | No | Smallest file is 39.6 GiB | n/a | n/a |
| Qwen2.5-72B-Instruct | No | Smallest file is 41.0 GiB | n/a | n/a |
| Qwen3-Next-80B-A3B-Instruct | No | Smallest file is 45.1 GiB | n/a | n/a |
| GLM-4.5-Air | No | Smallest file is 68.0 GiB | n/a | n/a |
| Qwen3-235B-A22B | No | Smallest file is 132.4 GiB | n/a | n/a |
llama.cpp GGUF files, one sequence, fp16 KV cache, 8,192 tokens. Mixture-of-experts models keep all weights resident, so their file size sets the memory need, not their active parameters. Estimates, not measurements on a Mac.
Where the 11 GiB comes from
Apple does not publish a per-size rule. Its Metal talk gives 21 GB for the GPU on a 32 GB machine and 48 GB on a 64 GB one, which is two thirds and three quarters. This page uses two thirds up to 32 GB and three quarters above that. Some newer 128 GB machines report about 84%, so on those you get more than this page assumes. The limit can be raised with sudo sysctl iogpu.wired_limit_mb=… on recent macOS; raising it leaves less for the system.
The catalog's own Apple records use a different reserve: M2 Max 64GB reserves 10 GiB and M4 Max 128GB reserves 12 GiB, which assumes you have raised the limit. Their numbers will be higher than the 64 and 128 GB rows here.
Data sources
- Apple silicon unified memory and LLM sizing (LLMS-Explorer) — Quotes Apple's Metal Compute talk: 32 GB Mac gives the GPU 21 GB, 64 GB gives 48 GB. Newer 128 GB M4 Max logs show about 84%, so the share is not a constant.
- recommendedMaxWorkingSetSize (Apple Developer Forums) — Discussion of the default GPU working-set limit on Apple silicon.
- Model file sizes and architectures — catalog dataset 2026-10-02.1; GGUF bytes come from the Hugging Face repos listed on each model page