Your rig
M3 Ultra · 256 GB · 256 GB
256 GB unified · 800 GB/s · 192.0 GB addressable by the GPU
Which one should I actually run?
Best models for your M3 Ultra · 256 GB
What do you want to do? Each row is the best-ranked model on that use case's shortlist that fits at Q5_K_M and 4K. The ordering is editorial; the sizes and speeds are computed. How the shortlists work.
All
General assistantCodingReasoning & mathsWritingVisionAgents & tool useTranslationLow latencyLong context
| Recommendation | Model | Why | Quant · total | Tok/s est. | |
|---|---|---|---|---|---|
| Best overall | Qwen3.8 27B New this week Alibaba · Apache 2.0 |
Best capability per gigabyte: a current-generation dense 27B with vision and 262K context. | Q5_K_M · 19.2 GB | 24 | Runs great |
| Best for coding | Ling 3.0 Flash 124B-A5B New inclusionAI · MIT |
SWE-bench Pro 56.6 claimed at 5B active — the 128 GB-class coding pick if you cannot fit DeepSeek V4. | Q5_K_M · 82.5 GB | 62 | Runs great |
| Best reasoning | Ling 3.0 Flash 124B-A5B New inclusionAI · MIT |
AIME 93 claimed at 5B active; hybrid attention keeps long reasoning traces cheap. | Q5_K_M · 82.5 GB | 62 | 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. | Q5_K_M · 48.5 GB | 9.6 | Runs great |
| Best vision | Qwen3.8 27B New this week Alibaba · Apache 2.0 |
Image and video in, OSWorld 84 — the strongest local vision-language model at 24 GB. | Q5_K_M · 19.2 GB | 24 | Runs great |
| Best for agents | Ling 3.0 Flash 124B-A5B New inclusionAI · MIT |
Agentic workflows are what Ling 3.0 was trained for; MIT, 5B active. | Q5_K_M · 82.5 GB | 62 | Runs great |
| Best translation | Qwen3.8 27B New this week Alibaba · Apache 2.0 |
119 languages, and Qwen translations read as idiomatic rather than literal. | Q5_K_M · 19.2 GB | 24 | Runs great |
| Fastest good model | Qwen3.6 35B-A3B Alibaba · Apache 2.0 |
3.3B active per token: the fastest model that still competes with dense 27Bs. | Q5_K_M · 24.4 GB | 96 | Runs great |
| Best long context | Qwen3.8 27B New this week Alibaba · Apache 2.0 |
262K native, and only 16 of 64 blocks keep a KV cache — long context is cheap here. | Q5_K_M · 27.0 GB at 128K context |
24 | Runs great |
Everything that fits
Runs great
67 of 76 · fits with headroom for a longer prompt| Model | Params | Quant | Weights | +KV 4K | Total | VRAM headroom | Tok/s | Runtime | |
|---|---|---|---|---|---|---|---|---|---|
| Qwen3 235B-A22B | 235B MoE / 22B act | Q5_K_M | 155.1 GB | 0.73 GB | 156.5 GB |
35.5 GB free
|
14 | MLX | Run |
| Qwen3.5 122B-A10B | 125B MoE / 10B act | Q5_K_M | 82.5 GB | 0.09 GB | 83.2 GB |
108.8 GB free
|
32 | MLX | Run |
| Nemotron 3 Super 120B-A12B | 124B MoE / 12B act | Q5_K_M | 81.8 GB | 0.03 GB | 82.5 GB |
109.5 GB free
|
26 | MLX | Run |
| Ling 3.0 Flash 124B-A5B New | 124B MoE / 5.1B act | Q5_K_M | 81.8 GB | 0.03 GB | 82.5 GB |
109.5 GB free
|
62 | MLX | Run |
| Mistral Small 4 119B-A6B | 119B MoE / 6.5B act | Q5_K_M | 78.5 GB | 0.09 GB | 79.2 GB |
112.8 GB free
|
49 | MLX | Run |
| gpt-oss 120B | 117B MoE / 5.1B act | MXFP4 | 60.5 GB | 0.15 GB | 61.3 GB |
130.7 GB free
|
79 | MLX | Run |
| Llama 4 Scout 109B-A17B | 109B MoE / 17B act | Q5_K_M | 71.9 GB | 0.75 GB | 73.3 GB |
118.7 GB free
|
19 | MLX | Run |
| Llama 3.3 70B Instruct | 70.6B | Q5_K_M | 46.6 GB | 1.25 GB | 48.5 GB |
143.5 GB free
|
9.6 | MLX | Run |
| DeepSeek-R1-Distill-Llama 70B | 70.6B | Q5_K_M | 46.6 GB | 1.25 GB | 48.5 GB |
143.5 GB free
|
9.6 | MLX | Run |
| Qwen3.6 35B-A3B | 35.9B MoE / 3.3B act | Q5_K_M | 23.7 GB | 0.08 GB | 24.4 GB |
167.6 GB free
|
96 | MLX | Run |
| Ornith 1.5 35B-A3B New this week | 35.9B MoE / 3.3B act | Q5_K_M | 23.7 GB | 0.08 GB | 24.4 GB |
167.6 GB free
|
96 | MLX | Run |
| LLM-jp 4 33B Thinking New this week | 33.2B | Q5_K_M | 21.9 GB | 1.00 GB | 23.5 GB |
168.5 GB free
|
20 | MLX | Run |
| Qwen3 32B | 32.8B | Q5_K_M | 21.7 GB | 1.00 GB | 23.3 GB |
168.7 GB free
|
21 | MLX | Run |
| DeepSeek-R1-Distill-Qwen 32B | 32.8B | Q5_K_M | 21.7 GB | 1.00 GB | 23.3 GB |
168.7 GB free
|
21 | MLX | Run |
| Qwen2.5-Coder 32B | 32.8B | Q5_K_M | 21.7 GB | 1.00 GB | 23.3 GB |
168.7 GB free
|
21 | MLX | Run |
| Olmo 3.1 32B Instruct | 32.2B | Q5_K_M | 21.3 GB | 1.00 GB | 22.9 GB |
169.1 GB free
|
21 | MLX | Run |
| Nemotron 3.5 Lightning 30B-A3B New | 31.6B MoE / 3.2B act | Q5_K_M | 20.9 GB | 0.02 GB | 21.5 GB |
170.5 GB free
|
99 | MLX | Run |
| Gemma 4 31B | 31.3B | Q5_K_M | 20.7 GB | 1.41 GB | 22.7 GB |
169.3 GB free
|
22 | MLX | Run |
| GLM-4.7-Flash 30B-A3B | 31.2B MoE / 3B act | Q5_K_M | 20.6 GB | 0.21 GB | 21.4 GB |
170.6 GB free
|
105 | MLX | Run |
| Qwen3 30B-A3B | 30.5B MoE / 3.3B act | Q5_K_M | 20.1 GB | 0.38 GB | 21.1 GB |
170.9 GB free
|
96 | MLX | Run |
| Qwen3 Coder 30B-A3B | 30.5B MoE / 3.3B act | Q5_K_M | 20.1 GB | 0.38 GB | 21.1 GB |
170.9 GB free
|
96 | MLX | Run |
| Muse Glimmer 30B New | 29.8B | Q5_K_M | 19.7 GB | 0.13 GB | 20.4 GB |
171.6 GB free
|
23 | MLX | Run |
| Granite 4.1 30B | 28.9B | Q5_K_M | 19.1 GB | 1.00 GB | 20.7 GB |
171.3 GB free
|
23 | MLX | Run |
| Qwen3.8 27B New this week | 27.8B | Q5_K_M | 18.4 GB | 0.25 GB | 19.2 GB |
172.8 GB free
|
24 | MLX | Run |
| Qwen3.6 27B | 27.8B | Q5_K_M | 18.4 GB | 0.25 GB | 19.2 GB |
172.8 GB free
|
24 | MLX | Run |
| Gemma 3 27B | 27.4B | Q5_K_M | 18.1 GB | 0.72 GB | 19.4 GB |
172.6 GB free
|
25 | MLX | Run |
| Gemma 4 26B-A4B | 26.5B MoE / 3.8B act | Q5_K_M | 17.5 GB | 0.35 GB | 18.4 GB |
173.6 GB free
|
83 | MLX | Run |
| Devstral Small 2 24B | 24B | Q5_K_M | 15.8 GB | 0.63 GB | 17.1 GB |
174.9 GB free
|
28 | MLX | Run |
| Mistral Small 3.2 24B | 23.6B | Q5_K_M | 15.6 GB | 0.63 GB | 16.8 GB |
175.2 GB free
|
29 | MLX | Run |
| gpt-oss 20B | 20.9B MoE / 3.6B act | MXFP4 | 10.8 GB | 0.10 GB | 11.5 GB |
180.5 GB free
|
112 | MLX | Run |
| Qwen3 14B | 14.8B | Q5_K_M | 9.8 GB | 0.63 GB | 11.0 GB |
181.0 GB free
|
46 | MLX | Run |
| DeepSeek-R1-Distill-Qwen 14B | 14.8B | Q5_K_M | 9.8 GB | 0.75 GB | 11.1 GB |
180.9 GB free
|
46 | MLX | Run |
| Qwen2.5-Coder 14B | 14.8B | Q5_K_M | 9.8 GB | 0.75 GB | 11.1 GB |
180.9 GB free
|
46 | MLX | Run |
| Phi-4 14B | 14.7B | Q5_K_M | 9.7 GB | 0.78 GB | 11.1 GB |
180.9 GB free
|
46 | MLX | Run |
| Ministral 3 14B | 13.9B | Q5_K_M | 9.2 GB | 0.63 GB | 10.4 GB |
181.6 GB free
|
49 | MLX | Run |
| Gemma 3 12B | 12.2B | Q5_K_M | 8.1 GB | 0.56 GB | 9.2 GB |
182.8 GB free
|
56 | MLX | Run |
| Mistral NeMo 12B | 12.2B | Q5_K_M | 8.1 GB | 0.63 GB | 9.3 GB |
182.7 GB free
|
56 | MLX | Run |
| Gemma 4 12B | 12B | Q5_K_M | 7.9 GB | 0.56 GB | 9.1 GB |
182.9 GB free
|
56 | MLX | Run |
| Qwen3.5 9B | 9.65B | Q5_K_M | 6.4 GB | 0.13 GB | 7.1 GB |
184.9 GB free
|
70 | MLX | Run |
| Ornith 1.5 9B New this week | 9.41B | Q5_K_M | 6.2 GB | 0.13 GB | 6.9 GB |
185.1 GB free
|
72 | MLX | Run |
| Ministral 3 8B | 8.92B | Q5_K_M | 5.9 GB | 0.53 GB | 7.0 GB |
185.0 GB free
|
76 | MLX | Run |
| Granite 4.1 8B | 8.79B | Q5_K_M | 5.8 GB | 0.63 GB | 7.0 GB |
185.0 GB free
|
77 | MLX | Run |
| LFM2.5 8B-A1B | 8.47B MoE / 1.5B act | Q5_K_M | 5.6 GB | 0.05 GB | 6.2 GB |
185.8 GB free
|
211 | MLX | Run |
| Fara 7B | 8.29B | Q5_K_M | 5.5 GB | 0.22 GB | 6.3 GB |
185.7 GB free
|
82 | MLX | Run |
| Qwen3 8B | 8.19B | Q5_K_M | 5.4 GB | 0.56 GB | 6.6 GB |
185.4 GB free
|
83 | MLX | Run |
| Llama 3.1 8B Instruct | 8.03B | Q5_K_M | 5.3 GB | 0.50 GB | 6.4 GB |
185.6 GB free
|
84 | MLX | Run |
| Gemma 4 E4B | 8.0B | Q5_K_M | 5.3 GB | 0.09 GB | 6.0 GB |
186.0 GB free
|
85 | MLX | Run |
| Ling 3.0 Tiny 7.9B-A1.3B New | 7.9B MoE / 1.3B act | Q5_K_M | 5.2 GB | 0.03 GB | 5.8 GB |
186.2 GB free
|
243 | MLX | Run |
| DeepSeek-R1-Distill-Qwen 7B | 7.62B | Q5_K_M | 5.0 GB | 0.22 GB | 5.8 GB |
186.2 GB free
|
89 | MLX | Run |
| Qwen2.5-Coder 7B | 7.62B | Q5_K_M | 5.0 GB | 0.22 GB | 5.8 GB |
186.2 GB free
|
89 | MLX | Run |
| Olmo 3 7B Instruct | 7.3B | Q5_K_M | 4.8 GB | 2.00 GB | 7.4 GB |
184.6 GB free
|
93 | MLX | Run |
| Mistral 7B Instruct v0.3 | 7.25B | Q5_K_M | 4.8 GB | 0.50 GB | 5.9 GB |
186.1 GB free
|
93 | MLX | Run |
| Gemma 4 E2B | 5.1B | Q5_K_M | 3.4 GB | 0.04 GB | 4.0 GB |
188.0 GB free
|
133 | MLX | Run |
| Qwen3.5 4B | 4.66B | Q5_K_M | 3.1 GB | 0.13 GB | 3.8 GB |
188.2 GB free
|
145 | MLX | Run |
| Gemma 3 4B | 4.3B | Q5_K_M | 2.8 GB | 0.20 GB | 3.6 GB |
188.4 GB free
|
157 | MLX | Run |
| Qwen3 4B | 4.02B | Q5_K_M | 2.7 GB | 0.56 GB | 3.8 GB |
188.2 GB free
|
168 | MLX | Run |
| Ministral 3 3B | 3.85B | Q5_K_M | 2.5 GB | 0.41 GB | 3.5 GB |
188.5 GB free
|
176 | MLX | Run |
| Phi-4-mini 3.8B | 3.84B | Q5_K_M | 2.5 GB | 0.50 GB | 3.6 GB |
188.4 GB free
|
176 | MLX | Run |
| Granite 4.1 3B | 3.4B | Q5_K_M | 2.2 GB | 0.31 GB | 3.2 GB |
188.8 GB free
|
199 | MLX | Run |
| Llama 3.2 3B Instruct | 3.21B | Q5_K_M | 2.1 GB | 0.44 GB | 3.2 GB |
188.8 GB free
|
211 | MLX | Run |
| LFM2.5 2.6B New | 2.7B | Q5_K_M | 1.8 GB | 0.06 GB | 2.4 GB |
189.6 GB free
|
251 | MLX | Run |
| Qwen3.5 2B | 2.27B | Q5_K_M | 1.5 GB | 0.05 GB | 2.1 GB |
189.9 GB free
|
298 | MLX | Run |
| Qwen3 1.7B | 1.72B | Q5_K_M | 1.1 GB | 0.44 GB | 2.2 GB |
189.8 GB free
|
394 | MLX | Run |
| Llama 3.2 1B Instruct | 1.24B | Q5_K_M | 0.8 GB | 0.13 GB | 1.5 GB |
190.5 GB free
|
546 | MLX | Run |
| Gemma 3 1B | 1.0B | Q5_K_M | 0.7 GB | 0.03 GB | 1.3 GB |
190.7 GB free
|
677 | MLX | Run |
| Qwen3.5 0.8B | 0.87B | Q5_K_M | 0.6 GB | 0.05 GB | 1.2 GB |
190.8 GB free
|
778 | MLX | Run |
| Qwen3 0.6B | 0.6B | Q5_K_M | 0.4 GB | 0.44 GB | 1.4 GB |
190.6 GB free
|
1129 | MLX | Run |
Runs — tight
1 model · fits at 4K, drop context or quant to go longer| Model | Params | Quant | Weights | +KV 4K | Total | VRAM headroom | Tok/s | Runtime | |
|---|---|---|---|---|---|---|---|---|---|
| DeepSeek V4 Flash 284B-A13B New | 284B MoE / 13B act | Q5_K_M | 187.5 GB | 0.19 GB | 188.3 GB |
3.7 GB free
|
24 | MLX | Run |
Won't fit in VRAM
8 models · some are possible with CPU offload| Model | Params | Quant | Weights | +KV 4K | Total | Over budget | Tok/s | Runtime | |
|---|---|---|---|---|---|---|---|---|---|
| Kimi K3 2.8T-A104B New | 2780B MoE / 104B act | Q5_K_M | 1835.0 GB | 0.11 GB | 1835.7 GB | +1643.7 GB | ~3.0 | MLX | Why |
| Qwen3.8 2.4T-A95B New | 2446B MoE / 95B act | Q5_K_M | 1614.5 GB | 0.36 GB | 1615.5 GB | +1423.5 GB | ~3.3 | MLX | Why |
| DeepSeek V4 Pro 1.6T-A49B New | 1650B MoE / 49B act | Q5_K_M | 1089.1 GB | 0.27 GB | 1090.0 GB | +898.0 GB | ~6.4 | MLX | Why |
| Kimi K2.6 1T-A32B | 1027B MoE / 32B act | Q5_K_M | 677.9 GB | 0.27 GB | 678.8 GB | +486.8 GB | ~9.9 | MLX | Why |
| GLM-5.2 744B-A40B | 753B MoE / 40B act | Q5_K_M | 497.0 GB | 0.34 GB | 498.0 GB | +306.0 GB | ~7.9 | MLX | Why |
| DeepSeek-R1 671B | 671B MoE / 37B act | Q5_K_M | 442.9 GB | 0.27 GB | 443.8 GB | +251.8 GB | ~8.5 | MLX | Why |
| Qwen3.5 397B-A17B | 403B MoE / 17B act | Q5_K_M | 266.0 GB | 0.12 GB | 266.7 GB | +74.7 GB | ~19 | MLX | Why |
| Ornith 1.5 397B-A17B New this week | 397B MoE / 17B act | Q5_K_M | 262.0 GB | 0.12 GB | 262.8 GB | +70.8 GB | ~19 | MLX | Why |
Totals = quantised weights + f16 KV cache at 4K tokens + 0.6 GB runtime overhead, against 192.0 GB usable unified memory. Tokens per second are estimated from 800 GB/s peak memory bandwidth at batch 1 — not measured. Read the method.