What AI models run on the Mac M3 Ultra 512GB?
The Mac M3 Ultra 512GB has 512 GB of unified memory (819 GB/s). About 384 GB is usable for a model. 278 of 285 open models we track fit at 8K context.
278models fit (8K ctx)
278still fit at 32K ctx
1.03Tlargest that fits
819 GB/smemory bandwidth
Models that fit, largest first
| Model | Params | Best quant (8K) | Needs | 32K context | Max tok/s (est.) | |
|---|---|---|---|---|---|---|
| Kimi K2.5 | 1.03T (33B act.) | Q2_K | 349 GB | Q2_K | 68 | details |
| Kimi K2.7 Code | 1.03T (33B act.) | Q2_K | 349 GB | Q2_K | 68 | details |
| Kimi K2.6 | 1.03T (33B act.) | Q2_K | 335 GB | Q2_K | 70 | details |
| Kimi K2 Thinking | 1.03T (33B act.) | Q2_K | 350 GB | Q2_K | 68 | details |
| Kimi K2 Instruct 0905 | 1.03T (33B act.) | Q2_K | 350 GB | Q2_K | 68 | details |
| Kimi K2 Instruct | 1.03T (33B act.) | Q2_K | 349 GB | Q2_K | 69 | details |
| Kimi K2 Base | 1.03T (33B act.) | Q2_K | 360 GB | Q2_K | 66 | details |
| MiMo V2.6 Pro RL | 1.02T (45B act.) | Q2_K | 362 GB | Q2_K | 49 | details |
| MiMo V2.5 Pro | 1.02T (44B act.) | Q2_K | 361 GB | Q2_K | 50 | details |
| Ling 1T | 1000B (51B act.) | Q2_K | 342 GB | Q2_K | 44 | details |
| Inkling | 952B (19B act.) | Q2_K | 335 GB | Q2_K | — | details |
| Muse Spark Open 816B A42B | 822B (42B act.) | Q2_K | 285 GB | Q2_K | 53 | details |
| Hy4 preview | 780B (40B act.) | Q2_K | 274 GB | Q2_K | 54 | details |
| DeepSeek V4.1 Flash 1017 | 763B (188B act.) | Q2_K | 268 GB | Q2_K | 12 | details |
| DeepSeek V4.1 Flash | 763B (188B act.) | Q2_K | 268 GB | Q2_K | 12 | details |
| GLM 5 | 754B (42B act.) | Q3_K_M | 337 GB | Q3_K_M | 40 | details |
| GLM 5.1 | 754B (42B act.) | Q3_K_M | 317 GB | Q3_K_M | 43 | details |
| GLM 5.3 | 753B (42B act.) | Q3_K_M | 344 GB | Q2_K | 40 | details |
| GLM 5.2 | 753B (42B act.) | Q3_K_M | 321 GB | Q3_K_M | 43 | details |
| DeepSeek V3.2 Exp | 685B (41B act.) | Q3_K_M | 313 GB | Q3_K_M | 41 | details |
| DeepSeek V3.2 | 685B (41B act.) | Q3_K_M | 300 GB | Q3_K_M | 43 | details |
| DeepSeek V3.1 | 685B (40B act.) | Q3_K_M | 300 GB | Q3_K_M | 44 | details |
| DeepSeek V3 0324 | 685B (40B act.) | Q3_K_M | 298 GB | Q3_K_M | 44 | details |
| DeepSeek R1 0528 | 685B (40B act.) | Q3_K_M | 299 GB | Q3_K_M | 44 | details |
| DeepSeek R1 | 684B (40B act.) | Q3_K_M | 298 GB | Q3_K_M | 44 | details |
| DeepSeek V3 | 684B (40B act.) | Q3_K_M | 298 GB | Q3_K_M | 44 | details |
| Inkling 2 Small | 594B (19B act.) | Q3_K_M | 263 GB | Q3_K_M | 90 | details |
| Ling 3.1 flash 1019 | 572B (25B act.) | Q4_K_M | 335 GB | Q4_K_M | 52 | details |
| Ling 3.1 flash | 572B (25B act.) | Q4_K_M | 335 GB | Q4_K_M | 52 | details |
| LongCat Flash Chat | 562B (562B act.) | Q4_K_M | 321 GB | Q4_K_M | 2.4 | details |
| MiniMax M1 40k | 456B (48B act.) | Q5_K_M | 306 GB | Q5_K_M | 24 | details |
| MiniMax Text 01 | 456B (48B act.) | Q5_K_M | 306 GB | Q5_K_M | 24 | details |
| MiniMax M3 | 427B (20B act.) | Q6_K | 333 GB | Q6_K | 50 | details |
| Llama 4 Maverick 17B 128E Instruct | 402B (18B act.) | Q6_K | 308 GB | Q6_K | 55 | details |
| GLM 4.5 | 358B (36B act.) | Q6_K | 278 GB | Q6_K | 28 | details |
| GLM 4.7 | 358B (36B act.) | Q6_K | 278 GB | Q6_K | 28 | details |
| GLM 4.6 | 357B (34B act.) | Q6_K | 277 GB | Q6_K | 29 | details |
| GLM 5.3 Flash | 321B (18B act.) | Q8_0 | 318 GB | Q8_0 | 42 | details |
| Nex N2.6 | 321B (18B act.) | Q8_0 | 319 GB | Q8_0 | 42 | details |
| step3 | 321B | Q6_K | 276 GB | Q5_K_M | 3.1 | details |
| MiMo V2.6 Flash RL | 311B (6.2B act.) | Q8_0 | 309 GB | Q8_0 | — | details |
| MiMo V2.5 | 311B (17B act.) | Q8_0 | 308 GB | Q8_0 | 44 | details |
| MiMo V2.6 Flash MOPD | 311B (6.2B act.) | Q8_0 | 309 GB | Q8_0 | — | details |
| MiMo V2 Flash | 310B (16B act.) | Q8_0 | 308 GB | Q8_0 | 47 | details |
| DeepSeek V4 Flash Vision Exp | 305B (15B act.) | Q8_0 | 303 GB | Q8_0 | 51 | details |
| DeepSeek V4 Flash 0731 | 304B (27B act.) | Q8_0 | 302 GB | Q8_0 | 28 | details |
| DeepSeek V4 Flash DSpark | 304B (27B act.) | Q8_0 | 302 GB | Q8_0 | 28 | details |
| Hy3 | 299B (21B act.) | Q8_0 | 299 GB | Q8_0 | 37 | details |
| Hy3 preview | 299B (21B act.) | Q8_0 | 299 GB | Q8_0 | 37 | details |
| DeepSeek V4 Flash | 291B (14B act.) | Q8_0 | 289 GB | Q8_0 | 55 | details |
| ERNIE 5.5 Flash Base | 286B (5.7B act.) | Q8_0 | 284 GB | Q8_0 | — | details |
| Intern S1 | 241B (28B act.) | Q8_0 | 235 GB | Q8_0 | 28 | details |
| DeepSeek V2 Chat | 236B (21B act.) | Q8_0 | 234 GB | Q8_0 | 36 | details |
| MiniMax M2.5 | 229B (11B act.) | Q8_0 | 229 GB | Q8_0 | 70 | details |
| MiniMax M2.7 | 229B (11B act.) | Q8_0 | 229 GB | Q8_0 | 70 | details |
| MiniMax M2 | 229B (11B act.) | Q8_0 | 229 GB | Q8_0 | 70 | details |
| MiniMax M2.1 | 229B (11B act.) | Q8_0 | 229 GB | Q8_0 | 70 | details |
| MiniMax M3.1 Flash | 220B (4.4B act.) | Q8_0 | 220 GB | Q8_0 | — | details |
| Step 3.7 Flash | 201B | Q8_0 | 203 GB | Q8_0 | 3.8 | details |
| Step 3.5 Flash | 199B | Q8_0 | 199 GB | Q8_0 | 3.9 | details |
| Qwen3.8 Flash Next | 180B (62B act.) | F16 | 331 GB | F16 | 6.8 | details |
| Ling 3.0 flash Sante | 127B (5.6B act.) | F16 | 239 GB | F16 | 73 | details |
| Ling 3.0 flash | 127B (5.6B act.) | F16 | 239 GB | F16 | 73 | details |
| Ornith 1.5 122B | 125B (13B act.) | F16 | 233 GB | F16 | 32 | details |
| gpt oss 120b | 117B (5.8B act.) | F16 | 61.9 GB | F16 | 255 | details |
| gpt oss safeguard 120b | 117B (5.8B act.) | F16 | 219 GB | F16 | 71 | details |
| GLM 4.5 Air | 110B (15B act.) | F16 | 208 GB | F16 | 27 | details |
| Llama 4 Scout 17B 16E Instruct | 109B (18B act.) | F16 | 203 GB | F16 | 23 | details |
| Llama 4 Scout 17B 16E | 109B (18B act.) | F16 | 205 GB | F16 | 23 | details |
| GLM 4.5V | 108B (14B act.) | F16 | 203 GB | F16 | 29 | details |
| GLM 4.6V | 108B (14B act.) | F16 | 201 GB | F16 | 29 | details |
| Hunyuan A13B Instruct | 80B (13B act.) | F16 | 151 GB | F16 | 31 | details |
| Kimi Dev 72B | 73B | F16 | 139 GB | F16 | 5.6 | details |
| Llama 3.3 70B Instruct | 71B | F16 | 135 GB | F16 | 5.8 | details |
| DeepSeek R1 Distill Llama 70B | 71B | F16 | 135 GB | F16 | 5.8 | details |
| Llama 3.1 70B | 71B | F16 | 135 GB | F16 | 5.8 | details |
| Kimi Linear 48B A3B Instruct | 49B (3.0B act.) | F16 | 92.5 GB | F16 | 136 | details |
| Nous Hermes 2 Mixtral 8x7B DPO | 47B (13B act.) | F16 | 88.5 GB | F16 | 32 | details |
| falcon 40b | 42B | F16 | 79.4 GB | F16 | 9.8 | details |
| flint | 40B (1.3B act.) | F16 | 75.1 GB | F16 | 328 | details |
| Maxwell V1 Flash | 37B (2.3B act.) | F16 | 69.7 GB | F16 | 180 | details |
| Hermes 4.3 36B | 36B | F16 | 70.0 GB | F16 | 11 | details |
| Seed OSS 36B Instruct | 36B | F16 | 70.0 GB | F16 | 11 | details |
| Qwen3.6 35B A3B | 36B (3.0B act.) | F16 | 67.2 GB | F16 | 138 | details |
| Ornith 1.5 35B A3B | 36B (3.0B act.) | F16 | 67.2 GB | F16 | 138 | details |
| Qwen3.5 35B A3B | 36B (3.0B act.) | F16 | 65.6 GB | F16 | 141 | details |
| Axion 1.0 35B | 35B (3.9B act.) | F16 | 65.6 GB | F16 | 106 | details |
| Qwen AgentWorld 35B A3B | 35B (3.0B act.) | F16 | 65.6 GB | F16 | 136 | details |
| Yi 34B 200K | 34B | F16 | 66.5 GB | F16 | 12 | details |
| Yi 34B | 34B | F16 | 66.5 GB | F16 | 12 | details |
| Yi 34B Chat | 34B | F16 | 66.5 GB | F16 | 12 | details |
| Yi 1.5 34B Chat | 34B | F16 | 66.5 GB | F16 | 12 | details |
| Nous Hermes 2 Yi 34B | 34B | F16 | 66.5 GB | F16 | 12 | details |
| DeepSeek R1 Distill Qwen 32B | 33B | F16 | 63.7 GB | F16 | 12 | details |
| Qwen3 32B | 33B | F16 | 63.7 GB | F16 | 12 | details |
| GLM Z1 32B 0414 | 33B | F16 | 61.8 GB | F16 | 13 | details |
| GLM 4 32B 0414 | 33B | F16 | 61.8 GB | F16 | 13 | details |
| granite 4.0 h small | 32B (8.8B act.) | F16 | 60.7 GB | F16 | 46 | details |
| gemma 4 31B it | 31B | F16 | 60.1 GB | F16 | 13 | details |
| GLM 4.7 Flash | 31B (4.6B act.) | F16 | 56.9 GB | F16 | 93 | details |
| Qwen3 Coder 30B A3B Instruct | 31B (3.0B act.) | F16 | 58.3 GB | F16 | 136 | details |
| North Mini Code 1.0 | 30B (3.3B act.) | F16 | 58.2 GB | F16 | 124 | details |
| Hy Memory 30B A3B | 30B (3.0B act.) | F16 | 57.4 GB | F16 | 137 | details |
| Muse Glimmer 30B | 30B | F16 | 52.9 GB | F16 | 15 | details |
| granite 4.2 30b | 29B | F16 | 57.1 GB | F16 | 14 | details |
| Qwen3.8 27B | 28B | F16 | 52.3 GB | F16 | 15 | details |
| Qwen3.6 27B | 28B | F16 | 51.5 GB | F16 | 15 | details |
| Qwen3.5 27B | 28B | F16 | 51.5 GB | F16 | 15 | details |
| gemma 3 27b it | 27B | F16 | 52.2 GB | F16 | 15 | details |
| Arete 27B | 27B | F16 | 51.5 GB | F16 | 15 | details |
| MiniMax H3 Image PE | 27B | F16 | 52.3 GB | F16 | 15 | details |
| Trinity Mini | 26B (2.0B act.) | F16 | 49.7 GB | F16 | 208 | details |
| GLiDE 27B | 26B | F16 | 50.0 GB | F16 | 16 | details |
| diffusiongemma 26B A4B it | 26B (4.0B act.) | F16 | 48.5 GB | F16 | 105 | details |
| gemma 4 26B A4B it | 26B (4.0B act.) | F16 | 48.4 GB | F16 | 105 | details |
| LFM2.5 24B A2B | 24B (2.0B act.) | F16 | 90.1 GB | F16 | 102 | details |
| ERNIE 4.5 21B A3B PT | 22B | F16 | 41.7 GB | F16 | 19 | details |
| ERNIE 4.5 21B A3B Thinking | 22B | F16 | 41.7 GB | F16 | 19 | details |
| gpt oss safeguard 20b | 22B (4.8B act.) | F16 | 13.8 GB | F16 | 267 | details |
| gpt oss 20b | 21B (4.2B act.) | F16 | 13.8 GB | F16 | 296 | details |
| Groot 19B A3B 261023 | 19B (3.0B act.) | F16 | 36.4 GB | F16 | 140 | details |
| Kimi VL A3B Thinking 2506 | 16B (3.0B act.) | F16 | 60.4 GB | F16 | 70 | details |
| Kimi VL A3B Instruct | 16B (3.0B act.) | F16 | 31.5 GB | F16 | 137 | details |
| Kimi VL A3B Thinking | 16B (3.0B act.) | F16 | 31.5 GB | F16 | 137 | details |
| deepseek moe 16b base | 16B (2.8B act.) | F16 | 32.8 GB | F16 | 145 | details |
| LLaDA2.1 mini | 16B (1.4B act.) | F16 | 31.3 GB | F16 | 286 | details |
| Moonlight 16B A3B Instruct | 16B (3.0B act.) | F16 | 30.7 GB | F16 | 137 | details |
| DeepSeek Coder V2 Lite Instruct | 16B (2.7B act.) | F16 | 30.1 GB | F16 | 154 | details |
| DeepSeek V2 Lite | 16B (2.7B act.) | F16 | 30.1 GB | F16 | 154 | details |
| DeepSeek V2 Lite Chat | 16B (2.7B act.) | F16 | 30.1 GB | F16 | 154 | details |
| Qwen2.5 Coder 14B Instruct | 15B | F16 | 29.7 GB | F16 | 28 | details |
| DeepSeek R1 Distill Qwen 14B | 15B | F16 | 29.7 GB | F16 | 28 | details |
| Qwen3 14B | 15B | F16 | 29.4 GB | F16 | 28 | details |
| phi 4 | 15B | F16 | 29.5 GB | F16 | 28 | details |
| gemma 3 12b it | 12B | F16 | 23.6 GB | F16 | 35 | details |
| Llama Guard 4 12B | 12B | F16 | 24.6 GB | F16 | 34 | details |
| Llama 3.2 11B Vision Instruct | 11B | F16 | 20.1 GB | F16 | 42 | details |
| Llama 3.2 11B Vision | 11B | F16 | 21.7 GB | F16 | 38 | details |
| GLM 4.6V Flash | 10B | F16 | 18.5 GB | F16 | 44 | details |
| GLM 4.1V 9B Thinking | 10B | F16 | 18.5 GB | F16 | 44 | details |
| Step3 VL 10B | 10B | F16 | 32.3 GB | F16 | 25 | details |
| SenseNova Loop 9B Preview | 9.7B | F16 | 17.8 GB | F16 | 46 | details |
| Qwen3.5 9B | 9.7B | F16 | 17.8 GB | F16 | 46 | details |
| Ornith 1.5 9B | 9.7B | F16 | 17.8 GB | F16 | 46 | details |
| Qwen3.5 9B Base | 9.7B | F16 | 17.8 GB | F16 | 46 | details |
| MiMo V2.6 Distill Qwen 9B | 9.4B | F16 | 17.8 GB | F16 | 46 | details |
| glm 4 9b | 9.4B | F16 | 23.2 GB | F16 | 44 | details |
| internlm3 8b instruct | 8.8B | F16 | 17.4 GB | F16 | 47 | details |
| granite 4.2 8b | 8.8B | F16 | 18.2 GB | F16 | 47 | details |
| granite 4.1 8b | 8.8B | F16 | 18.2 GB | F16 | 47 | details |
| Qwen3 VL 8B Instruct | 8.8B | F16 | 17.1 GB | F16 | 50 | details |
| Molmo2 8B | 8.7B | F16 | 17.1 GB | F16 | 50 | details |
| LFM2.5 8B A1B | 8.5B (1.0B act.) | F16 | 32.3 GB | F16 | 205 | details |
| Qwen2.5 VL 7B Instruct | 8.3B | F16 | 15.3 GB | F16 | 54 | details |
| UI TARS 1.5 7B | 8.3B | F16 | 15.3 GB | F16 | 54 | details |
| olmOCR 2 7B 1025 | 8.3B | F16 | 15.3 GB | F16 | 54 | details |
| olmOCR 7B 0225 preview | 8.3B | F16 | 16.6 GB | F16 | 49 | details |
| UI TARS 7B DPO | 8.3B | F16 | 15.3 GB | F16 | 54 | details |
| Qwen3 8B | 8.2B | F16 | 17.1 GB | F16 | 50 | details |
| DeepSeek R1 0528 Qwen3 8B | 8.2B | F16 | 17.1 GB | F16 | 50 | details |
| granite 3.3 8b instruct | 8.2B | F16 | 17.0 GB | F16 | 50 | details |
| granite 3.1 8b instruct | 8.2B | F16 | 17.0 GB | F16 | 50 | details |
| granite 3.2 8b instruct | 8.2B | F16 | 17.0 GB | F16 | 50 | details |
| Apertus 8B Instruct 2509 | 8.1B | F16 | 16.7 GB | F16 | 51 | details |
| Llama 3.1 8B Instruct | 8.0B | F16 | 16.6 GB | F16 | 51 | details |
| Llama 3.1 8B | 8.0B | F16 | 16.6 GB | F16 | 51 | details |
| Hermes 3 Llama 3.1 8B | 8.0B | F16 | 16.6 GB | F16 | 51 | details |
| DeepSeek R1 Distill Llama 8B | 8.0B | F16 | 31.6 GB | F16 | 25 | details |
| Meta Llama 3 8B Instruct | 8.0B | F16 | 16.6 GB | F16 | 51 | details |
| Meta Llama Guard 2 8B | 8.0B | F16 | 16.6 GB | F16 | 51 | details |
| Meta Llama 3.1 8B Instruct | 8.0B | F16 | 16.6 GB | F16 | 51 | details |
| Ling 3.0 tiny | 7.9B (1.4B act.) | F16 | 15.6 GB | F16 | 296 | details |
| MiMo 7B RL | 7.8B | F16 | 16.4 GB | F16 | 52 | details |
| internlm2 5 7b chat | 7.7B | F16 | 16.0 GB | F16 | 53 | details |
| internlm2 chat 7b | 7.7B | F16 | 16.0 GB | F16 | 53 | details |
| Qwen2.5 7B Instruct | 7.6B | F16 | 15.3 GB | F16 | 54 | details |
| Qwen2.5 Coder 7B Instruct | 7.6B | F16 | 15.3 GB | F16 | 54 | details |
| DeepSeek R1 Distill Qwen 7B | 7.6B | F16 | 15.3 GB | F16 | 54 | details |
| Falcon H1R 7B | 7.6B | F16 | 15.1 GB | F16 | 54 | details |
| Olmo 3 7B Instruct | 7.3B | F16 | 16.7 GB | F16 | 56 | details |
| Olmo 3 1025 7B | 7.3B | F16 | 16.7 GB | F16 | 56 | details |
| Olmo 3 7B Think | 7.3B | F16 | 16.7 GB | F16 | 56 | details |
| Mistral 7B v0.1 | 7.2B | F16 | 14.9 GB | F16 | 57 | details |
| Mistral 7B Instruct v0.2 | 7.2B | F16 | 14.9 GB | F16 | 57 | details |
| Mistral 7B Instruct v0.1 | 7.2B | F16 | 14.9 GB | F16 | 57 | details |
| falcon 7b | 7.2B | F16 | 18.4 GB | F16 | 57 | details |
| falcon 7b instruct | 7.2B | F16 | 18.4 GB | F16 | 57 | details |
| granite 4.0 h tiny | 6.9B (1.5B act.) | F16 | 13.6 GB | F16 | 279 | details |
| OLMoE 1B 7B 0924 | 6.9B (1.3B act.) | F16 | 14.4 GB | F16 | 319 | details |
| deepseek coder 7b instruct v1.5 | 6.9B | F16 | 17.2 GB | F16 | 59 | details |
| deepseek coder 6.7b instruct | 6.7B | F16 | 17.0 GB | F16 | 61 | details |
| deepseek coder 6.7b base | 6.7B | F16 | 17.0 GB | F16 | 61 | details |
| Llama 2 7b hf | 6.7B | F16 | 17.0 GB | F16 | 61 | details |
| Yi 6B 200K | 6.1B | F16 | 12.3 GB | F16 | 68 | details |
| Yi 6B | 6.1B | F16 | 12.3 GB | F16 | 68 | details |
| Molmo2 4B | 4.9B | F16 | 10.0 GB | F16 | 93 | details |
| WeVisOne VL 4B | 4.8B | F16 | 10.0 GB | F16 | 93 | details |
| Mage VL Cowork | 4.7B | F16 | 10.0 GB | F16 | 93 | details |
| Qianfan OCR | 4.7B | F16 | 10.1 GB | F16 | 93 | details |
| Qwen3.5 4B | 4.7B | F16 | 9.0 GB | F16 | 97 | details |
| S5 nano | 4.7B | F16 | 9.2 GB | F16 | 95 | details |
| LightOnOCR 3 4B | 4.5B | F16 | 9.0 GB | F16 | 97 | details |
| SingVL S1 4B | 4.5B | F16 | 9.0 GB | F16 | 97 | details |
| Intern Decision 4B | 4.5B | F16 | 9.0 GB | F16 | 97 | details |
| Qwen3 VL 4B Instruct | 4.4B | F16 | 9.3 GB | F16 | 102 | details |
| gemma 3 4b it | 4.3B | F16 | 8.4 GB | F16 | 105 | details |
| medgemma 1.5 4b it | 4.3B | F16 | 8.4 GB | F16 | 105 | details |
| medgemma 4b it | 4.3B | F16 | 8.4 GB | F16 | 105 | details |
| gemma 3 4b pt | 4.3B | F16 | 9.2 GB | F16 | 95 | details |
| Phi 3.5 vision instruct | 4.1B | F16 | 11.2 GB | F16 | 99 | details |
| Qwen3 4B | 4.0B | F16 | 9.3 GB | F16 | 102 | details |
| Qwen3 4B Instruct 2507 | 4.0B | F16 | 9.3 GB | F16 | 102 | details |
| granite vision 4.1 4b | 4.0B | F16 | 7.6 GB | F16 | 120 | details |
| Phi 4 mini instruct | 3.8B | F16 | 8.9 GB | F16 | 107 | details |
| Phi 3 mini 4k instruct | 3.8B | F16 | 10.6 GB | F16 | 107 | details |
| Phi 3.5 mini instruct | 3.8B | F16 | 10.6 GB | F16 | 107 | details |
| Phi 3 mini 128k instruct | 3.8B | F16 | 10.6 GB | F16 | 107 | details |
| Qwen2.5 VL 3B Instruct | 3.8B | F16 | 6.7 GB | F16 | 133 | details |
| granite 4.2 3b | 3.7B | F16 | 8.0 GB | F16 | 112 | details |
| granite 4.1 3b llmon preview | 3.7B | F16 | 8.0 GB | F16 | 112 | details |
| Raon 1 nano | 3.5B | F16 | 7.0 GB | F16 | 131 | details |
| granite 4.1 3b | 3.4B | F16 | 7.6 GB | F16 | 120 | details |
| granite 4.0 micro | 3.4B | F16 | 7.6 GB | F16 | 120 | details |
| DeepSeek OCR 2 | 3.4B (1.2B act.) | F16 | 6.6 GB | F16 | 396 | details |
| deepseek vl2 tiny | 3.4B (1.2B act.) | F16 | 7.4 GB | F16 | 348 | details |
| Unlimited OCR 2 | 3.3B (1.1B act.) | F16 | 6.6 GB | F16 | 405 | details |
| Unlimited OCR | 3.3B (1.1B act.) | F16 | 6.6 GB | F16 | 407 | details |
| DeepSeek OCR | 3.3B (1.1B act.) | F16 | 6.6 GB | F16 | 407 | details |
| granite 3.1 3b a800m instruct | 3.3B (883M act.) | F16 | 7.1 GB | F16 | 463 | details |
| Llama 3.2 3B Instruct | 3.2B | F16 | 13.5 GB | F16 | 64 | details |
| Llama 3.2 3B | 3.2B | F16 | 7.5 GB | F16 | 127 | details |
| Hermes 3 Llama 3.2 3B | 3.2B | F16 | 7.5 GB | F16 | 127 | details |
| d1 3B | 3.1B | F16 | 10.8 GB | F16 | 76 | details |
| LFM2.5 VL 3B | 3.1B | F16 | 10.8 GB | F16 | 76 | details |
| Qwen2.5 3B Instruct | 3.1B | F16 | 6.7 GB | F16 | 133 | details |
| SmolLM3 3B | 3.1B | F16 | 6.9 GB | F16 | 133 | details |
| SmolLM3 3B Base | 3.1B | F16 | 6.9 GB | F16 | 133 | details |
| phi 2 | 2.8B | F16 | 8.2 GB | F16 | 147 | details |
| LFM2.5 2.6B | 2.7B | F16 | 10.8 GB | F16 | 76 | details |
| granite 3.3 2b instruct | 2.5B | F16 | 5.8 GB | F16 | 162 | details |
| MiniCPM5 2B | 2.5B | F16 | 5.7 GB | F16 | 163 | details |
| WeVisOne VL 2B | 2.4B | F16 | 5.4 GB | F16 | 201 | details |
| Qwen3.5 2B | 2.3B | F16 | 4.5 GB | F16 | 217 | details |
| SmolVLM Instruct | 2.2B | F16 | 5.4 GB | F16 | 226 | details |
| Qwen3 1.7B | 2.0B | F16 | 4.8 GB | F16 | 238 | details |
| Qwen3.8 27B DSpine b16 | 2.0B | F16 | 4.8 GB | F16 | 202 | details |
| d1 2B | 1.9B | F16 | 7.2 GB | F16 | 118 | details |
| LFM3 VL 2B | 1.9B | F16 | 7.2 GB | F16 | 118 | details |
| DeepSeek R1 Distill Qwen 1.5B | 1.8B | F16 | 4.2 GB | F16 | 230 | details |
| LFM3 1.7B | 1.7B | F16 | 7.2 GB | F16 | 118 | details |
| LFM3 1.7B Base | 1.7B | F16 | 4.0 GB | F16 | 237 | details |
| SmolLM2 1.7B Instruct | 1.7B | F16 | 5.2 GB | F16 | 239 | details |
| SmolLM2 1.7B | 1.7B | F16 | 5.2 GB | F16 | 239 | details |
| SmolLM 1.7B | 1.7B | F16 | 5.2 GB | F16 | 239 | details |
| granite 4.0 1b | 1.6B | F16 | 4.3 GB | F16 | 251 | details |
| Qwen2.5 1.5B Instruct | 1.5B | F16 | 3.8 GB | F16 | 265 | details |
| OLMo 2 0425 1B | 1.5B | F16 | 4.4 GB | F16 | 275 | details |
| deepseek coder 1.3b instruct | 1.3B | F16 | 4.5 GB | F16 | 304 | details |
| Llama 3.2 1B | 1.2B | F16 | 3.2 GB | F16 | 331 | details |
| Llama 3.2 1B Instruct | 1.2B | F16 | 5.5 GB | F16 | 165 | details |
| LFM2.5 1.2B Instruct | 1.2B | F16 | 5.0 GB | F16 | 175 | details |
| HunyuanOCR | 1.1B | F16 | 3.5 GB | F16 | 366 | details |
| PaddleOCR VL 1.6 | 959M | F16 | 2.5 GB | F16 | 427 | details |
| Qwen3.5 0.8B | 873M | F16 | 2.4 GB | F16 | 540 | details |
| LightOnOCR 3 0.8B | 853M | F16 | 2.4 GB | F16 | 540 | details |
| FastVLM 0.5B | 759M | F16 | 2.2 GB | F16 | 540 | details |
| Qwen3 0.6B | 752M | F16 | 2.7 GB | F16 | 684 | details |
| GOT OCR2 0 | 716M | F16 | 2.8 GB | F16 | 572 | details |
| Nemotron 3.5 Parse | 631M | F16 | 1.9 GB | F16 | 627 | details |
| GOT OCR 2.0 hf | 561M | F16 | 2.5 GB | F16 | 731 | details |
| SmolVLM2 500M Video Instruct | 507M | F16 | 1.6 GB | F16 | 998 | details |
| SmolVLM 500M Instruct | 507M | F16 | 1.6 GB | F16 | 998 | details |
| Qwen2.5 0.5B Instruct | 494M | F16 | 1.7 GB | F16 | 829 | details |
| LFM2.5 VL 450M | 449M | F16 | 1.9 GB | F16 | 576 | details |
| SmolLM2 360M Instruct | 362M | F16 | 1.5 GB | F16 | 1129 | details |
| SmolLM2 360M | 362M | F16 | 1.5 GB | F16 | 1129 | details |
| LFM2.5 350M | 354M | F16 | 1.9 GB | F16 | 576 | details |
| LFM2 350M | 354M | F16 | 1.4 GB | F16 | 1155 | details |
Just too big at 4-bit
These need up to 35% more memory than the Mac M3 Ultra 512GB has at Q4_K_M and 8K context. A smaller quant or shorter context may work.
| Model | Q4_K_M needs | Short by |
|---|---|---|
| DeepSeek V3.2 Exp | 391.5 GB | 7.5 GB |
| GLM 5 | 426.1 GB | 42.1 GB |
| GLM 5.3 | 430.4 GB | 46.4 GB |
| GLM 5.1 | 434.1 GB | 50.1 GB |
| GLM 5.2 | 435.4 GB | 51.4 GB |
| DeepSeek V4.1 Flash 1017 | 435.8 GB | 51.8 GB |
| DeepSeek V4.1 Flash | 435.8 GB | 51.8 GB |
| Hy4 preview | 445.5 GB | 61.5 GB |
| Muse Spark Open 816B A42B | 469.6 GB | 85.6 GB |
Questions
What is the biggest model that runs on the Mac M3 Ultra 512GB?
Of the models we track, the largest that fits at 8K context is Kimi K2.5 (1.03T) at Q2_K, needing about 349 GB.
How much of the Mac M3 Ultra 512GB's memory can a model use?
About 384 GB of 512 GB. macOS lets the GPU use about 75% of unified memory by default; this can be raised with the iogpu.wired_limit_mb setting.
Spec source: Apple specifications. Model data: Hugging Face Hub API. Speed is a bandwidth ceiling (bandwidth ÷ bytes read per token), not a benchmark. Methodology · Report an error.