Which LLMs Fit on a 256 GB Mac?

GLM-5.3-Flash 320B (MoE) ranks first and is also the largest model that fits, at 215.1 GB with room to spare. MiMo-V2.6-Flash 309B needs 207.6 GB, DeepSeek V4 Flash 284B 190.7 GB and Qwen3 235B (MoE A22B) 159.5 GB.

All of these are mixture of experts models that run 13B to 22B parameters per token, so they write text at a usable pace on an M5 Ultra. The largest dense model that fits is Mistral Medium 3.5 128B, at 89.2 GB. A dense model reads all its weights for every token, so the big MoE models are the reason to get this much memory.

GLM-5.3-Flash barely grows at 32K. It needs 215.5 GB there, since its KV cache takes only 0.6 GB. Qwen3 235B keeps a full cache on all 94 of its layers, so its cache is 6.4 GB at 32K. It still fits, at 164.7 GB.

For 64K or 128K, open the model finder from the card below the table and change the context there. Long prompts take a while to read on any Mac, even when the model fits.

At Q8_0, the limit drops to the 180B class. Qwen3.8-Flash-Next 180B (MoE) is the largest that fits, at 210.3 GB, and Qwen3 235B needs 275.8 GB. Mistral Medium 3.5 128B fits at Q8_0 even with 32K, at 162.3 GB.

The 300B class only fits at Q4_K_M. A big model at Q4_K_M usually gives better answers than a smaller one at Q8_0, so most people with this much memory run the largest MoE model they can at Q4_K_M.

K2 Horizon 375B A23B (MoE) misses by 5.9 GB at Q4_K_M. At Q3_K_M it needs 204.4 GB and fits. Llama 3.1 405B is similar: 276.5 GB at Q4_K_M and 223.0 GB at Q3_K_M. It's dense, though, so it runs slowly on any Mac.

The 671B models stay out of reach even at Q3_K_M, where DeepSeek R1 needs 362.4 GB. They're on the 512 GB Mac list.

Apple sells 256 GB with the M5 Ultra in the Mac Studio. The Ultra has the most memory bandwidth of any Mac, and bandwidth is what sets token speed for models this size.

macOS keeps part of the memory away from the GPU by default. On macOS Sonoma and later, this command lets the GPU use 248 GB until the next restart:

sudo sysctl iogpu.wired_limit_mb=253952

Keep an eye on memory pressure in Activity Monitor. If the Mac starts swapping to disk, lower the limit or pick a smaller model.