RunsOnMy ModelsGPUs & MacsAPI pricesVRAM calculatorNew

What AI models run on the Mac M3 Pro 36GB?

The Mac M3 Pro 36GB has 36 GB of unified memory (150 GB/s). About 24.1 GB is usable for a model. 202 of 285 open models we track fit at 8K context.

202models fit (8K ctx)
201still fit at 32K ctx
49Blargest that fits
150 GB/smemory bandwidth

Models that fit, largest first

ModelParamsBest quant (8K)Needs32K contextMax tok/s (est.)
Kimi Linear 48B A3B Instruct49B (3.0B act.)Q2_K17.3 GBQ2_K140details
Nous Hermes 2 Mixtral 8x7B DPO47B (13B act.)Q2_K17.6 GBQ2_K31details
falcon 40b42BQ3_K_M20.2 GBQ2_K7.5details
flint40B (1.3B act.)Q3_K_M19.0 GBQ3_K_M246details
Maxwell V1 Flash37B (2.3B act.)Q3_K_M17.9 GBQ3_K_M135details
Hermes 4.3 36B36BQ3_K_M19.1 GBQ2_K8.5details
Seed OSS 36B Instruct36BQ3_K_M19.1 GBQ2_K8.5details
Qwen3.6 35B A3B36B (3.0B act.)Q3_K_M17.0 GBQ3_K_M105details
Ornith 1.5 35B A3B36B (3.0B act.)Q4_K_M21.4 GBQ3_K_M82details
Qwen3.5 35B A3B36B (3.0B act.)Q4_K_M21.5 GBQ3_K_M82details
Axion 1.0 35B35B (3.9B act.)Q4_K_M20.7 GBQ4_K_M64details
Qwen AgentWorld 35B A3B35B (3.0B act.)Q4_K_M21.6 GBQ3_K_M78details
Yi 34B 200K34BQ4_K_M21.6 GBQ2_K7.3details
Yi 34B34BQ4_K_M21.6 GBQ2_K7.3details
Yi 34B Chat34BQ4_K_M21.6 GBQ2_K7.3details
Yi 1.5 34B Chat34BQ4_K_M21.6 GBQ2_K7.3details
Nous Hermes 2 Yi 34B34BQ3_K_M18.4 GBQ2_K8.7details
DeepSeek R1 Distill Qwen 32B33BQ4_K_M21.2 GBQ2_K7.6details
Qwen3 32B33BQ4_K_M21.1 GBQ2_K7.6details
GLM Z1 32B 041433BQ4_K_M19.5 GBQ4_K_M7.6details
GLM 4 32B 041433BQ4_K_M19.5 GBQ4_K_M7.6details
granite 4.0 h small32B (8.8B act.)Q4_K_M18.9 GBQ4_K_M28details
gemma 4 31B it31BQ4_K_M20.0 GBQ3_K_M8.2details
GLM 4.7 Flash31B (4.6B act.)Q5_K_M21.0 GBQ4_K_M47details
Qwen3 Coder 30B A3B Instruct31B (3.0B act.)Q5_K_M21.7 GBQ4_K_M70details
North Mini Code 1.030B (3.3B act.)Q4_K_M19.3 GBQ4_K_M72details
Hy Memory 30B A3B30B (3.0B act.)Q5_K_M21.3 GBQ4_K_M70details
Muse Glimmer 30B30BQ5_K_M18.8 GBQ5_K_M7.8details
granite 4.2 30b29BQ4_K_M19.1 GBQ2_K8.5details
Qwen3.8 27B28BQ5_K_M19.9 GBQ5_K_M7.5details
Qwen3.6 27B28BQ5_K_M19.5 GBQ5_K_M7.7details
Qwen3.5 27B28BQ5_K_M19.6 GBQ5_K_M7.6details
gemma 3 27b it27BQ5_K_M19.9 GBQ4_K_M7.8details
Arete 27B27BQ5_K_M20.6 GBQ4_K_M7.3details
MiniMax H3 Image PE27BQ5_K_M19.3 GBQ5_K_M7.8details
Trinity Mini26B (2.0B act.)Q6_K21.0 GBQ6_K93details
GLiDE 27B26BQ6_K21.3 GBQ5_K_M7.0details
diffusiongemma 26B A4B it26B (4.0B act.)Q5_K_M19.2 GBQ5_K_M51details
gemma 4 26B A4B it26B (4.0B act.)Q5_K_M21.1 GBQ4_K_M46details
LFM2.5 24B A2B24B (2.0B act.)Q6_K19.1 GBQ6_K91details
ERNIE 4.5 21B A3B PT22BQ6_K17.8 GBQ6_K8.3details
ERNIE 4.5 21B A3B Thinking22BQ6_K17.8 GBQ6_K8.3details
gpt oss safeguard 20b22B (4.8B act.)F1613.8 GBF1649details
gpt oss 20b21B (4.2B act.)F1613.8 GBF1654details
Groot 19B A3B 26102319B (3.0B act.)Q8_019.9 GBQ8_048details
Kimi VL A3B Thinking 250616B (3.0B act.)Q8_016.8 GBQ8_048details
Kimi VL A3B Instruct16B (3.0B act.)Q8_016.8 GBQ8_048details
Kimi VL A3B Thinking16B (3.0B act.)Q8_017.2 GBQ8_047details
deepseek moe 16b base16B (2.8B act.)Q8_018.6 GBQ6_K50details
LLaDA2.1 mini16B (1.4B act.)Q8_017.1 GBQ8_099details
Moonlight 16B A3B Instruct16B (3.0B act.)Q8_016.8 GBQ8_047details
DeepSeek Coder V2 Lite Instruct16B (2.7B act.)Q8_016.4 GBQ8_053details
DeepSeek V2 Lite16B (2.7B act.)Q8_016.4 GBQ8_053details
DeepSeek V2 Lite Chat16B (2.7B act.)Q8_016.4 GBQ8_053details
Qwen2.5 Coder 14B Instruct15BQ8_016.8 GBQ8_09.6details
DeepSeek R1 Distill Qwen 14B15BQ8_016.8 GBQ8_09.6details
Qwen3 14B15BQ8_016.6 GBQ8_09.6details
phi 415BQ8_016.7 GBQ8_09.6details
gemma 3 12b it12BQ8_013.4 GBQ8_012details
Llama Guard 4 12B12BQ8_014.2 GBQ8_012details
Llama 3.2 11B Vision Instruct11BF1620.1 GBQ8_07.7details
Llama 3.2 11B Vision11BQ8_012.4 GBQ8_013details
GLM 4.6V Flash10BF1618.5 GBF168.0details
GLM 4.1V 9B Thinking10BF1618.5 GBF168.0details
Step3 VL 10B10BQ8_09.9 GBQ8_017details
SenseNova Loop 9B Preview9.7BF1617.8 GBF168.4details
Qwen3.5 9B9.7BF1617.8 GBF168.4details
Ornith 1.5 9B9.7BF1617.8 GBF168.4details
Qwen3.5 9B Base9.7BF1617.8 GBF168.4details
MiMo V2.6 Distill Qwen 9B9.4BF1617.8 GBF168.4details
glm 4 9b9.4BQ8_015.0 GBno15details
internlm3 8b instruct8.8BF1617.4 GBF168.5details
granite 4.2 8b8.8BF1618.2 GBQ8_08.5details
granite 4.1 8b8.8BF1618.2 GBQ8_08.5details
Qwen3 VL 8B Instruct8.8BF1617.1 GBF169.2details
Molmo2 8B8.7BF1617.1 GBF169.2details
LFM2.5 8B A1B8.5B (1.0B act.)Q8_09.1 GBQ8_0141details
Qwen2.5 VL 7B Instruct8.3BF1615.3 GBF169.8details
UI TARS 1.5 7B8.3BF1615.3 GBF169.8details
olmOCR 2 7B 10258.3BF1615.3 GBF169.8details
olmOCR 7B 0225 preview8.3BF1616.6 GBF169.0details
UI TARS 7B DPO8.3BF1615.3 GBF169.8details
Qwen3 8B8.2BF1617.1 GBF169.2details
DeepSeek R1 0528 Qwen3 8B8.2BF1617.1 GBF169.2details
granite 3.3 8b instruct8.2BF1617.0 GBF169.2details
granite 3.1 8b instruct8.2BF1617.0 GBF169.2details
granite 3.2 8b instruct8.2BF1617.0 GBF169.2details
Apertus 8B Instruct 25098.1BF1616.7 GBF169.3details
Llama 3.1 8B Instruct8.0BF1616.6 GBF169.3details
Llama 3.1 8B8.0BF1616.6 GBF169.3details
Hermes 3 Llama 3.1 8B8.0BF1616.6 GBF169.3details
DeepSeek R1 Distill Llama 8B8.0BQ8_09.6 GBQ8_018details
Meta Llama 3 8B Instruct8.0BF1616.6 GBF169.3details
Meta Llama Guard 2 8B8.0BF1616.6 GBF169.3details
Meta Llama 3.1 8B Instruct8.0BF1616.6 GBF169.3details
Ling 3.0 tiny7.9B (1.4B act.)F1615.6 GBF1654details
MiMo 7B RL7.8BF1616.4 GBF169.6details
internlm2 5 7b chat7.7BF1616.0 GBF169.7details
internlm2 chat 7b7.7BF1616.0 GBF169.7details
Qwen2.5 7B Instruct7.6BF1615.3 GBF169.8details
Qwen2.5 Coder 7B Instruct7.6BF1615.3 GBF169.8details
DeepSeek R1 Distill Qwen 7B7.6BF1615.3 GBF169.8details
Falcon H1R 7B7.6BF1615.1 GBF169.9details
Olmo 3 7B Instruct7.3BF1616.7 GBF1610details
Olmo 3 1025 7B7.3BF1616.7 GBF1610details
Olmo 3 7B Think7.3BF1616.7 GBF1610details
Mistral 7B v0.17.2BF1614.9 GBF1610details
Mistral 7B Instruct v0.27.2BF1614.9 GBF1610details
Mistral 7B Instruct v0.17.2BF1614.9 GBF1610details
falcon 7b7.2BF1618.4 GBQ5_K_M10details
falcon 7b instruct7.2BF1618.4 GBQ5_K_M10details
granite 4.0 h tiny6.9B (1.5B act.)F1613.6 GBF1651details
OLMoE 1B 7B 09246.9B (1.3B act.)F1614.4 GBF1658details
deepseek coder 7b instruct v1.56.9BF1617.2 GBQ6_K11details
deepseek coder 6.7b instruct6.7BF1617.0 GBQ6_K11details
deepseek coder 6.7b base6.7BF1617.0 GBQ6_K11details
Llama 2 7b hf6.7BF1617.0 GBQ6_K11details
Yi 6B 200K6.1BF1612.3 GBF1612details
Yi 6B6.1BF1612.3 GBF1612details
Molmo2 4B4.9BF1610.0 GBF1617details
WeVisOne VL 4B4.8BF1610.0 GBF1617details
Mage VL Cowork4.7BF1610.0 GBF1617details
Qianfan OCR4.7BF1610.1 GBF1617details
Qwen3.5 4B4.7BF169.0 GBF1618details
S5 nano4.7BF169.2 GBF1617details
LightOnOCR 3 4B4.5BF169.0 GBF1618details
SingVL S1 4B4.5BF169.0 GBF1618details
Intern Decision 4B4.5BF169.0 GBF1618details
Qwen3 VL 4B Instruct4.4BF169.3 GBF1619details
gemma 3 4b it4.3BF168.4 GBF1619details
medgemma 1.5 4b it4.3BF168.4 GBF1619details
medgemma 4b it4.3BF168.4 GBF1619details
gemma 3 4b pt4.3BF169.2 GBF1617details
Phi 3.5 vision instruct4.1BF1611.2 GBF1618details
Qwen3 4B4.0BF169.3 GBF1619details
Qwen3 4B Instruct 25074.0BF169.3 GBF1619details
granite vision 4.1 4b4.0BF167.6 GBF1622details
Phi 4 mini instruct3.8BF168.9 GBF1620details
Phi 3 mini 4k instruct3.8BF1610.6 GBF1620details
Phi 3.5 mini instruct3.8BF1610.6 GBF1620details
Phi 3 mini 128k instruct3.8BF1610.6 GBF1620details
Qwen2.5 VL 3B Instruct3.8BF166.7 GBF1624details
granite 4.2 3b3.7BF168.0 GBF1620details
granite 4.1 3b llmon preview3.7BF168.0 GBF1620details
Raon 1 nano3.5BF167.0 GBF1624details
granite 4.1 3b3.4BF167.6 GBF1622details
granite 4.0 micro3.4BF167.6 GBF1622details
DeepSeek OCR 23.4B (1.2B act.)F166.6 GBF1672details
deepseek vl2 tiny3.4B (1.2B act.)F167.4 GBF1664details
Unlimited OCR 23.3B (1.1B act.)F166.6 GBF1674details
Unlimited OCR3.3B (1.1B act.)F166.6 GBF1675details
DeepSeek OCR3.3B (1.1B act.)F166.6 GBF1675details
granite 3.1 3b a800m instruct3.3B (883M act.)F167.1 GBF1685details
Llama 3.2 3B Instruct3.2BF1613.5 GBF1612details
Llama 3.2 3B3.2BF167.5 GBF1623details
Hermes 3 Llama 3.2 3B3.2BF167.5 GBF1623details
d1 3B3.1BF1610.8 GBF1614details
LFM2.5 VL 3B3.1BF1610.8 GBF1614details
Qwen2.5 3B Instruct3.1BF166.7 GBF1624details
SmolLM3 3B3.1BF166.9 GBF1624details
SmolLM3 3B Base3.1BF166.9 GBF1624details
phi 22.8BF168.2 GBF1627details
LFM2.5 2.6B2.7BF1610.8 GBF1614details
granite 3.3 2b instruct2.5BF165.8 GBF1630details
MiniCPM5 2B2.5BF165.7 GBF1630details
WeVisOne VL 2B2.4BF165.4 GBF1637details
Qwen3.5 2B2.3BF164.5 GBF1640details
SmolVLM Instruct2.2BF165.4 GBF1641details
Qwen3 1.7B2.0BF164.8 GBF1644details
Qwen3.8 27B DSpine b162.0BF164.8 GBF1637details
d1 2B1.9BF167.2 GBF1622details
LFM3 VL 2B1.9BF167.2 GBF1622details
DeepSeek R1 Distill Qwen 1.5B1.8BF164.2 GBF1642details
LFM3 1.7B1.7BF167.2 GBF1622details
LFM3 1.7B Base1.7BF164.0 GBF1643details
SmolLM2 1.7B Instruct1.7BF165.2 GBF1644details
SmolLM2 1.7B1.7BF165.2 GBF1644details
SmolLM 1.7B1.7BF165.2 GBF1644details
granite 4.0 1b1.6BF164.3 GBF1646details
Qwen2.5 1.5B Instruct1.5BF163.8 GBF1649details
OLMo 2 0425 1B1.5BF164.4 GBF1650details
deepseek coder 1.3b instruct1.3BF164.5 GBF1656details
Llama 3.2 1B1.2BF163.2 GBF1661details
Llama 3.2 1B Instruct1.2BF165.5 GBF1630details
LFM2.5 1.2B Instruct1.2BF165.0 GBF1632details
HunyuanOCR1.1BF163.5 GBF1667details
PaddleOCR VL 1.6959MF162.5 GBF1678details
Qwen3.5 0.8B873MF162.4 GBF1699details
LightOnOCR 3 0.8B853MF162.4 GBF1699details
FastVLM 0.5B759MF162.2 GBF1699details
Qwen3 0.6B752MF162.7 GBF16125details
GOT OCR2 0716MF162.8 GBF16105details
Nemotron 3.5 Parse631MF161.9 GBF16115details
GOT OCR 2.0 hf561MF162.5 GBF16134details
SmolVLM2 500M Video Instruct507MF161.6 GBF16183details
SmolVLM 500M Instruct507MF161.6 GBF16183details
Qwen2.5 0.5B Instruct494MF161.7 GBF16152details
LFM2.5 VL 450M449MF161.9 GBF16105details
SmolLM2 360M Instruct362MF161.5 GBF16207details
SmolLM2 360M362MF161.5 GBF16207details
LFM2.5 350M354MF161.9 GBF16105details
LFM2 350M354MF161.4 GBF16212details

Just too big at 4-bit

These need up to 35% more memory than the Mac M3 Pro 36GB has at Q4_K_M and 8K context. A smaller quant or shorter context may work.

ModelQ4_K_M needsShort by
falcon 40b25.2 GB1.1 GB
Nous Hermes 2 Mixtral 8x7B DPO28.0 GB3.8 GB
Kimi Linear 48B A3B Instruct28.9 GB4.8 GB

Questions

What is the biggest model that runs on the Mac M3 Pro 36GB?

Of the models we track, the largest that fits at 8K context is Kimi Linear 48B A3B Instruct (49B) at Q2_K, needing about 17.3 GB.

How much of the Mac M3 Pro 36GB's memory can a model use?

About 24 GB of 36 GB. macOS lets the GPU use about 67% 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.