Best LLM for 256 GB unified memory · 2026

Best LLM for a 256 GB Mac: what fits, and what to run

Updated 8 Oct 2026 1 new this month

macOS lets the GPU wire down about 192.0 GB of a 256 GB Mac's unified memory; the rest stays with the system and your apps. At Q4_K_M with an 8K context, 70 of 80 models in the catalogue fit — 69 with headroom, 1 tightly. The pick for most people is DeepSeek V4 Flash 284B-A13B: Frontier-adjacent at 13B active — Q4 on a 256 GB machine, Q3 squeezed into 192 GB. Speeds are estimated for an M3 Ultra · 256 GB; the fit verdicts are the same for every machine in the class.

Machines in this class M3 Ultra · 256 GB

Which one should I run?

One pick per job, from the editorial shortlist, checked to fit at Q4_K_M and 8K. Each label links the full ranking for that job. How picks are chosen.

RecommendationModelWhyQuant · totalTok/s est.
Best overall DeepSeek V4 Flash 284B-A13B
DeepSeek · MIT
Frontier-adjacent at 13B active — Q4 on a 256 GB machine, Q3 squeezed into 192 GB. Q4_K_M · 160.7 GB 29 Runs great
Best for coding DeepSeek V4 Flash 284B-A13B
DeepSeek · MIT
Frontier coding at 13B active for 192 GB+ machines. Q4_K_M · 160.7 GB 29 Runs great
Best reasoning DeepSeek V4 Flash 284B-A13B
DeepSeek · MIT
Most of V4 Pro's reasoning at 13B active. Q4_K_M · 160.7 GB 29 Runs great
Best for writing Llama 3.3 70B Instruct
Meta · Llama 3.3 Community
The creative-writing favourite: consistent voice, takes direction, handles fiction framing. Q4_K_M · 42.8 GB 11 Runs great
Best vision Qwen3.8-Flash-Next 180B-A6B
Alibaba · Qwen Community 1.0
Qwen3.8 vision on a 192 GB machine at 6B active. Q4_K_M · 102.0 GB 62 Runs great
Best for agents DeepSeek V4 Flash 284B-A13B
DeepSeek · MIT
Long agent trajectories at 13B active. Q4_K_M · 160.7 GB 29 Runs great
Best translation Qwen3.8 27B
Alibaba · Apache 2.0
119 languages, and Qwen translations read as idiomatic rather than literal. Q4_K_M · 16.7 GB 29 Runs great
Fastest good model Qwen3.8-Flash-Next 180B-A6B
Alibaba · Qwen Community 1.0
6B active and a 12-layer KV cache: the fastest frontier-class model a 192 GB machine runs. Q4_K_M · 102.0 GB 62 Runs great
Best long context Qwen3.8-Flash-Next 180B-A6B
Alibaba · Qwen Community 1.0
262K with 2 KV heads on 12 of 48 blocks — the cheapest long context above 100B. Q4_K_M · 104.8 GB
at 128K context
62 Runs great

The long-context row is judged at 128K: Qwen3.8-Flash-Next 180B-A6B holds that window on a 256 GB Mac at Q4_K_M with 3.00 GB of KV cache — that is why it can differ from the best overall pick.

Everything that fits a 256 GB Mac

Largest first. "Max context" is the longest window the Mac holds at that quantisation with an f16 cache; a q8_0 cache roughly doubles it.

ModelParamsQuantWeights+KV 8KTotalHeadroomMax contextTok/s est.
DeepSeek V4 Flash 284B-A13B 284B MoE Q4_K_M 159.7 GB 0.38 GB 160.7 GB 31.3 GB 671K 29 Runs great
Qwen3 235B-A22B 235B MoE Q4_K_M 132.1 GB 1.47 GB 134.2 GB 57.8 GB 128K 17 Runs great
Qwen3.8-Flash-Next 180B-A6B 180B MoE Q4_K_M 101.2 GB 0.19 GB 102.0 GB 90.0 GB 256K 62 Runs great
Qwen3.5 122B-A10B 125B MoE Q4_K_M 70.3 GB 0.19 GB 71.1 GB 120.9 GB 256K 37 Runs great
Nemotron 3 Super 120B-A12B 124B MoE Q4_K_M 69.7 GB 0.06 GB 70.4 GB 121.6 GB 256K 31 Runs great
Ling 3.0 Flash 124B-A5B 124B MoE Q4_K_M 69.7 GB 0.06 GB 70.4 GB 121.6 GB 256K 73 Runs great
Mistral Small 4 119B-A6B 119B MoE Q4_K_M 66.9 GB 0.18 GB 67.7 GB 124.3 GB 256K 57 Runs great
gpt-oss 120B 117B MoE MXFP4 60.5 GB 0.29 GB 61.4 GB 130.6 GB 128K 79 Runs great
Llama 4 Scout 109B-A17B 109B MoE Q4_K_M 61.3 GB 1.50 GB 63.4 GB 128.6 GB 1024K 22 Runs great
Llama 3.3 70B Instruct 70.6B Q4_K_M 39.7 GB 2.50 GB 42.8 GB 149.2 GB 128K 11 Runs great
DeepSeek-R1-Distill-Llama 70B 70.6B Q4_K_M 39.7 GB 2.50 GB 42.8 GB 149.2 GB 128K 11 Runs great
Qwen3.6 35B-A3B 35.9B MoE Q4_K_M 20.2 GB 0.16 GB 20.9 GB 171.1 GB 256K 112 Runs great
Ornith 1.5 35B-A3B 35.9B MoE Q4_K_M 20.2 GB 0.16 GB 20.9 GB 171.1 GB 256K 112 Runs great
LLM-jp 4 33B Thinking 33.2B Q4_K_M 18.7 GB 2.00 GB 21.3 GB 170.7 GB 64K 24 Runs great
Qwen3 32B 32.8B Q4_K_M 18.4 GB 2.00 GB 21.0 GB 171.0 GB 128K 24 Runs great
DeepSeek-R1-Distill-Qwen 32B 32.8B Q4_K_M 18.4 GB 2.00 GB 21.0 GB 171.0 GB 128K 24 Runs great
Qwen2.5-Coder 32B 32.8B Q4_K_M 18.4 GB 2.00 GB 21.0 GB 171.0 GB 128K 24 Runs great
Olmo 3.1 32B Instruct 32.2B Q4_K_M 18.1 GB 1.25 GB 20.0 GB 172.0 GB 64K 25 Runs great
Nemotron 3.5 Lightning 30B-A3B 31.6B MoE Q4_K_M 17.8 GB 0.05 GB 18.4 GB 173.6 GB 256K 116 Runs great
Gemma 4 31B 31.3B Q4_K_M 17.6 GB 2.03 GB 20.2 GB 171.8 GB 256K 25 Runs great
GLM-4.7-Flash 30B-A3B 31.2B MoE Q4_K_M 17.5 GB 0.41 GB 18.6 GB 173.4 GB 198K 124 Runs great
Qwen3 30B-A3B 30.5B MoE Q4_K_M 17.1 GB 0.75 GB 18.5 GB 173.5 GB 128K 112 Runs great
Qwen3 Coder 30B-A3B 30.5B MoE Q4_K_M 17.1 GB 0.75 GB 18.5 GB 173.5 GB 256K 112 Runs great
Muse Glimmer 30B 29.8B Q4_K_M 16.8 GB 0.18 GB 17.5 GB 174.5 GB 128K 27 Runs great
Granite 4.1 30B 28.9B Q4_K_M 16.3 GB 2.00 GB 18.9 GB 173.1 GB 128K 28 Runs great
Qwen3.8 27B 27.8B Q4_K_M 15.6 GB 0.50 GB 16.7 GB 175.3 GB 256K 29 Runs great
Qwen3.6 27B 27.8B Q4_K_M 15.6 GB 0.50 GB 16.7 GB 175.3 GB 256K 29 Runs great
Gemma 3 27B 27.4B Q4_K_M 15.4 GB 1.03 GB 17.0 GB 175.0 GB 128K 29 Runs great
Gemma 4 26B-A4B 26.5B MoE Q4_K_M 14.9 GB 0.51 GB 16.0 GB 176.0 GB 256K 98 Runs great
Devstral Small 2 24B 24B Q4_K_M 13.5 GB 1.25 GB 15.3 GB 176.7 GB 384K 33 Runs great
Mistral Small 3.2 24B 23.6B Q4_K_M 13.3 GB 1.25 GB 15.1 GB 176.9 GB 128K 34 Runs great
gpt-oss 20B 20.9B MoE MXFP4 10.8 GB 0.19 GB 11.6 GB 180.4 GB 128K 112 Runs great
Qwen3 14B 14.8B Q4_K_M 8.3 GB 1.25 GB 10.2 GB 181.8 GB 128K 54 Runs great
DeepSeek-R1-Distill-Qwen 14B 14.8B Q4_K_M 8.3 GB 1.50 GB 10.4 GB 181.6 GB 128K 54 Runs great
Qwen2.5-Coder 14B 14.8B Q4_K_M 8.3 GB 1.50 GB 10.4 GB 181.6 GB 128K 54 Runs great
Phi-4 14B 14.7B Q4_K_M 8.3 GB 1.56 GB 10.4 GB 181.6 GB 16K 54 Runs great
Ministral 3 14B 13.9B Q4_K_M 7.8 GB 1.25 GB 9.7 GB 182.3 GB 256K 57 Runs great
Gemma 3 12B 12.2B Q4_K_M 6.9 GB 0.81 GB 8.3 GB 183.7 GB 128K 65 Runs great
Mistral NeMo 12B 12.2B Q4_K_M 6.9 GB 1.25 GB 8.7 GB 183.3 GB 128K 65 Runs great
Gemma 4 12B 12B Q4_K_M 6.7 GB 0.81 GB 8.2 GB 183.8 GB 256K 66 Runs great
Qwen3.5 9B 9.65B Q4_K_M 5.4 GB 0.25 GB 6.3 GB 185.7 GB 256K 82 Runs great
Ornith 1.5 9B 9.41B Q4_K_M 5.3 GB 0.25 GB 6.1 GB 185.9 GB 256K 84 Runs great
Ministral 3 8B 8.92B Q4_K_M 5.0 GB 1.06 GB 6.7 GB 185.3 GB 256K 89 Runs great
Granite 4.1 8B 8.79B Q4_K_M 4.9 GB 1.25 GB 6.8 GB 185.2 GB 128K 90 Runs great
LFM2.5 8B-A1B 8.47B MoE Q4_K_M 4.8 GB 0.09 GB 5.5 GB 186.5 GB 125K 247 Runs great
Fara 7B 8.29B Q4_K_M 4.7 GB 0.44 GB 5.7 GB 186.3 GB 125K 96 Runs great
Qwen3 8B 8.19B Q4_K_M 4.6 GB 1.13 GB 6.3 GB 185.7 GB 128K 97 Runs great
Llama 3.1 8B Instruct 8.03B Q4_K_M 4.5 GB 1.00 GB 6.1 GB 185.9 GB 128K 99 Runs great
Gemma 4 E4B 8.0B Q4_K_M 4.5 GB 0.14 GB 5.2 GB 186.8 GB 128K 99 Runs great
Ling 3.0 Tiny 7.9B-A1.3B 7.9B MoE Q4_K_M 4.4 GB 0.05 GB 5.1 GB 186.9 GB 128K 285 Runs great
DeepSeek-R1-Distill-Qwen 7B 7.62B Q4_K_M 4.3 GB 0.44 GB 5.3 GB 186.7 GB 128K 104 Runs great
Qwen2.5-Coder 7B 7.62B Q4_K_M 4.3 GB 0.44 GB 5.3 GB 186.7 GB 128K 104 Runs great
Olmo 3 7B Instruct 7.3B Q4_K_M 4.1 GB 2.50 GB 7.2 GB 184.8 GB 64K 109 Runs great
Mistral 7B Instruct v0.3 7.25B Q4_K_M 4.1 GB 1.00 GB 5.7 GB 186.3 GB 32K 110 Runs great
Gemma 4 E2B 5.1B Q4_K_M 2.9 GB 0.07 GB 3.5 GB 188.5 GB 128K 156 Runs great
Qwen3.5 4B 4.66B Q4_K_M 2.6 GB 0.25 GB 3.5 GB 188.5 GB 256K 171 Runs great
Gemma 3 4B 4.3B Q4_K_M 2.4 GB 0.30 GB 3.3 GB 188.7 GB 128K 185 Runs great
Qwen3 4B 4.02B Q4_K_M 2.3 GB 1.13 GB 4.0 GB 188.0 GB 32K 198 Runs great
Ministral 3 3B 3.85B Q4_K_M 2.2 GB 0.81 GB 3.6 GB 188.4 GB 256K 207 Runs great
Phi-4-mini 3.8B 3.84B Q4_K_M 2.2 GB 1.00 GB 3.8 GB 188.2 GB 128K 207 Runs great
Granite 4.1 3B 3.4B Q4_K_M 1.9 GB 0.63 GB 3.1 GB 188.9 GB 128K 234 Runs great
Llama 3.2 3B Instruct 3.21B Q4_K_M 1.8 GB 0.88 GB 3.3 GB 188.7 GB 128K 248 Runs great
LFM2.5 2.6B 2.7B Q4_K_M 1.5 GB 0.13 GB 2.2 GB 189.8 GB 128K 294 Runs great
Qwen3.5 2B 2.27B Q4_K_M 1.3 GB 0.09 GB 2.0 GB 190.0 GB 256K 350 Runs great
Qwen3 1.7B 1.72B Q4_K_M 1.0 GB 0.88 GB 2.4 GB 189.6 GB 32K 462 Runs great
Llama 3.2 1B Instruct 1.24B Q4_K_M 0.7 GB 0.25 GB 1.5 GB 190.5 GB 128K 641 Runs great
Gemma 3 1B 1.0B Q4_K_M 0.6 GB 0.04 GB 1.2 GB 190.8 GB 32K 795 Runs great
Qwen3.5 0.8B 0.87B Q4_K_M 0.5 GB 0.09 GB 1.2 GB 190.8 GB 256K 914 Runs great
Qwen3 0.6B 0.6B Q4_K_M 0.3 GB 0.88 GB 1.8 GB 190.2 GB 32K 1325 Runs great
GLM-5.3-Flash 320B-A18B 321B MoE Q4_K_M 180.5 GB 0.09 GB 181.2 GB 10.8 GB 1015K 21 Tight

Just out of reach

The smallest models that do not fit. Unified memory has no slower pool to spill into, so there is no offload fallback: each needs a bigger machine, or a lower quantisation.

ModelParamsTotal @ Q4_K_MOver budgetNeeds
Ornith 1.5 397B-A17B 397B MoE 224.1 GB +32.1 GB 512 GB Mac
Qwen3.5 397B-A17B 403B MoE 227.4 GB +35.4 GB 512 GB Mac
DeepSeek V4.1 Flash 552B-A16B 552B MoE 311.0 GB +119.0 GB 512 GB Mac
DeepSeek-R1 671B 671B MoE 378.4 GB +186.4 GB no single Mac
GLM-5.2 744B-A40B 753B MoE 424.7 GB +232.7 GB no single Mac
GLM-5.3 744B-A40B 753B MoE 424.7 GB +232.7 GB no single Mac

Running it on a Mac

The GPU's share can be raised. The cap macOS puts on wired GPU memory is a setting, not a hardware limit: sudo sysctl iogpu.wired_limit_mb=… lifts it until the next reboot. Leave several gigabytes for macOS itself — every number on this page assumes the default share, so it under-promises rather than over.

MLX or GGUF. On Apple Silicon, MLX (via mlx-lm or LM Studio) is usually faster than llama.cpp's GGUF on the same model, and the community converts most popular releases within days. GGUF through Ollama or llama.cpp has the widest coverage and usually supports a new architecture first. The memory arithmetic is the same either way.

Speed is bandwidth. Every machine in this class holds the same models; what separates them is memory bandwidth (M3 Ultra 800 GB/s), and tokens per second scales with it.


← 192 GB Mac All memory classes 512 GB Mac → Adjust context, quant or cache →

Every total is quantised weights + f16 KV cache at 8K + 0.6 GB runtime overhead, against 192.0 GB usable. Tokens per second are estimated from memory bandwidth and never measured. The method.