On a 36 GB Mac, 12 catalog models run fully on the GPU at an 8,192-token context. The most capable is Qwen3.6-35B-A3B at Q4_K_M (needs ~23.9 GiB). Pick a smaller model or a lower quant for more headroom.
Figures assume a 36 GB unified-memory Mac (one pool shared by CPU and GPU), judged at an 8,192-token context. Speeds depend on the specific chip — see the GPU pages.
TL;DR
A 36 GB Mac is the first tier where the 30B class fits fully on the GPU at 4-bit sweet-spot quants — with modest headroom rather than a lot of it. The larger 40B+ mixture-of-experts models are borderline here rather than a given: macOS hands the GPU only part of the unified pool by default — roughly three quarters, and about two thirds on machines at 32 GB and below, so a 36 GB machine budgets against about 27. The list below is derived live from the fit engine.
The largest entries use most of what the GPU is allowed to take, and that pool is shared with macOS and your apps, so treat them as quiet-machine models. The quietly good option at this tier is a 30B-class model with slack left over — room for longer context, a browser full of tabs, and no memory pressure.
Mixture-of-experts entries deserve first look if responsiveness matters: they activate only a fraction of their parameters per token and, as a rule of thumb, feel quicker than dense models of similar total size.
Qwen3.6-35B-A3B at Q4_K_M · 36B params · needs ~23.9 GiB
One command to run it (llama.cpp):
llama-server -m Qwen3.6-35B-A3B-UD-Q4_K_M.gguf -c 8192 -ngl 999| Model | Sweet-spot quant | Fits in |
|---|---|---|
| Qwen3.6-35B-A3B Qwen | Q4_K_M Runs fully on GPU | ~23.9 GiB |
| Qwen3-Omni-30B-A3B-Instruct Qwen | Q4_K_M Runs fully on GPU | ~20.3 GiB |
| Qwen2.5-VL-32B-Instruct Qwen | Q4_K_M Runs fully on GPU | ~22.9 GiB |
| DeepSeek-R1-Distill-Qwen-32B deepseek-ai | Q4_K_M Runs fully on GPU | ~22.9 GiB |
| Qwen3-32B Qwen | Q4_K_M Runs fully on GPU | ~22.8 GiB |
| gemma-4-31B-it google | Q4_K_M Runs fully on GPU | ~21.3 GiB |
| GLM-4.7-Flash zai-org | Q6_K Runs fully on GPU | ~26.3 GiB |
| Qwen3-Coder-30B-A3B-Instruct Qwen | Q4_K_M Runs fully on GPU | ~20.3 GiB |
| Qwen3-30B-A3B-Instruct-2507 Qwen | Q4_K_M Runs fully on GPU | ~20.3 GiB |
| Qwen3.6-27B Qwen | Q4_K_M Runs fully on GPU | ~19.8 GiB |
| gemma-4-26B-A4B-it google | Q4_K_M Runs fully on GPU | ~18.4 GiB |
| Devstral-Small-2-24B-Instruct-2512 mistralai | Q4_K_M Runs fully on GPU | ~16.5 GiB |
Derived live from the fit engine + catalog at an 8,192-token context. "Fits in" is the modelled VRAM the sweet-spot quant needs (weights + KV cache + overhead). Speed depends on your specific card — check a GPU page or the calculator.
The 30B class, fully on the GPU at 4-bit sweet-spot quants — this is the first tier where that is a real fit rather than a squeeze. The larger 40B+ mixture-of-experts models are borderline here: some fit, some don't, and the list below is the answer for the ones we track. The quieter gain is headroom on the models you'd run daily.
Dense 70B-class models generally don't fit at healthy quants in 36 GB, and crushing one to very low bits usually gives worse output than a well-quantized 30B-class model. Larger MoE models are the more realistic reach at this tier — check the list below and the calculator for your exact context.
Less than the headline number. The pool is shared with macOS and your open apps, and macOS caps the GPU's share of unified memory by default. Advanced users can raise that limit, but leave the OS a working slice — a model that starves the system makes the whole machine unusable.
This page ranks by fit, not speed. A 36 GB configuration exists across different Apple Silicon tiers, and generation speed mostly tracks the chip's memory bandwidth — so two Macs with identical memory can generate at quite different rates. Check your specific chip's page for speed context.
Yes — the KV cache grows with context and shares the same pool as the weights and the OS. The verdicts here hold at the stated context; for long-document or long-chat use, verify your model and target context in the fit calculator first.
selected to compare · pick at least 2