RunsOnMy ModelsGPUs & MacsAPI pricesVRAM calculatorNew

What AI models run on the Mac M3 24GB?

The Mac M3 24GB has 24 GB of unified memory (100 GB/s). About 16.1 GB is usable for a model. 199 of 285 open models we track fit at 8K context.

199models fit (8K ctx)
181still fit at 32K ctx
40Blargest that fits
100 GB/smemory bandwidth

Models that fit, largest first

ModelParamsBest quant (8K)Needs32K contextMax tok/s (est.)
flint40B (1.3B act.)Q2_K14.8 GBQ2_K213details
Maxwell V1 Flash37B (2.3B act.)Q2_K14.0 GBQ2_K117details
Hermes 4.3 36B36BQ2_K15.4 GBno7.4details
Seed OSS 36B Instruct36BQ2_K15.4 GBno7.4details
Qwen3.6 35B A3B36B (3.0B act.)Q2_K13.6 GBQ2_K89details
Ornith 1.5 35B A3B36B (3.0B act.)Q2_K13.2 GBQ2_K92details
Qwen3.5 35B A3B36B (3.0B act.)Q2_K13.6 GBQ2_K89details
Axion 1.0 35B35B (3.9B act.)Q2_K13.3 GBQ2_K68details
Qwen AgentWorld 35B A3B35B (3.0B act.)Q2_K13.1 GBQ2_K89details
Yi 34B 200K34BQ2_K16.0 GBno6.9details
Yi 34B34BQ2_K16.0 GBno6.9details
Yi 34B Chat34BQ2_K16.0 GBno6.9details
Yi 1.5 34B Chat34BQ2_K14.3 GBno7.8details
Nous Hermes 2 Yi 34B34BQ2_K14.8 GBno7.5details
DeepSeek R1 Distill Qwen 32B33BQ2_K14.2 GBno8.1details
Qwen3 32B33BQ2_K14.1 GBno8.1details
GLM Z1 32B 041433BQ2_K12.6 GBQ2_K8.1details
GLM 4 32B 041433BQ2_K12.6 GBQ2_K8.1details
granite 4.0 h small32B (8.8B act.)Q2_K11.7 GBQ2_K31details
gemma 4 31B it31BQ2_K13.9 GBno8.5details
GLM 4.7 Flash31B (4.6B act.)Q2_K11.7 GBQ2_K60details
Qwen3 Coder 30B A3B Instruct31B (3.0B act.)Q2_K11.9 GBQ2_K90details
North Mini Code 1.030B (3.3B act.)Q2_K12.0 GBQ2_K81details
Hy Memory 30B A3B30B (3.0B act.)Q2_K11.9 GBQ2_K89details
Muse Glimmer 30B30BQ2_K11.4 GBQ2_K9.0details
granite 4.2 30b29BQ2_K12.7 GBno9.2details
Qwen3.8 27B28BQ3_K_M14.0 GBQ2_K7.4details
Qwen3.6 27B28BQ3_K_M14.0 GBQ2_K7.4details
Qwen3.5 27B28BQ3_K_M14.0 GBQ2_K7.4details
gemma 3 27b it27BQ3_K_M14.4 GBQ2_K7.4details
Arete 27B27BQ3_K_M13.6 GBQ2_K7.6details
MiniMax H3 Image PE27BQ3_K_M13.8 GBQ2_K7.5details
Trinity Mini26B (2.0B act.)Q3_K_M12.3 GBQ3_K_M110details
GLiDE 27B26BQ3_K_M13.2 GBQ2_K7.8details
diffusiongemma 26B A4B it26B (4.0B act.)Q3_K_M13.2 GBQ3_K_M51details
gemma 4 26B A4B it26B (4.0B act.)Q3_K_M13.3 GBQ3_K_M51details
LFM2.5 24B A2B24B (2.0B act.)Q4_K_M14.3 GBQ3_K_M83details
ERNIE 4.5 21B A3B PT22BQ4_K_M13.5 GBQ3_K_M7.5details
ERNIE 4.5 21B A3B Thinking22BQ4_K_M13.5 GBQ3_K_M7.5details
gpt oss safeguard 20b22B (4.8B act.)F1613.8 GBF1633details
gpt oss 20b21B (4.2B act.)F1613.8 GBF1636details
Groot 19B A3B 26102319B (3.0B act.)Q4_K_M12.7 GBQ4_K_M52details
Kimi VL A3B Thinking 250616B (3.0B act.)Q6_K13.5 GBQ6_K41details
Kimi VL A3B Instruct16B (3.0B act.)Q6_K14.2 GBQ5_K_M38details
Kimi VL A3B Thinking16B (3.0B act.)Q6_K13.5 GBQ6_K41details
deepseek moe 16b base16B (2.8B act.)Q5_K_M13.9 GBQ2_K47details
LLaDA2.1 mini16B (1.4B act.)Q6_K13.4 GBQ6_K85details
Moonlight 16B A3B Instruct16B (3.0B act.)Q6_K14.2 GBQ5_K_M37details
DeepSeek Coder V2 Lite Instruct16B (2.7B act.)Q6_K13.9 GBQ5_K_M42details
DeepSeek V2 Lite16B (2.7B act.)Q6_K13.9 GBQ5_K_M42details
DeepSeek V2 Lite Chat16B (2.7B act.)Q6_K13.9 GBQ5_K_M42details
Qwen2.5 Coder 14B Instruct15BQ6_K13.5 GBQ3_K_M8.3details
DeepSeek R1 Distill Qwen 14B15BQ6_K13.5 GBQ3_K_M8.2details
Qwen3 14B15BQ6_K13.2 GBQ4_K_M8.2details
phi 415BQ6_K13.4 GBQ3_K_M8.3details
gemma 3 12b it12BQ8_013.4 GBQ6_K8.0details
Llama Guard 4 12B12BQ8_014.2 GBQ4_K_M7.8details
Llama 3.2 11B Vision Instruct11BQ8_011.6 GBQ6_K9.6details
Llama 3.2 11B Vision11BQ8_012.4 GBQ6_K8.8details
GLM 4.6V Flash10BQ8_010.3 GBQ8_010details
GLM 4.1V 9B Thinking10BQ8_010.3 GBQ8_010details
Step3 VL 10B10BQ8_09.9 GBQ8_011details
SenseNova Loop 9B Preview9.7BQ8_010.0 GBQ8_010details
Qwen3.5 9B9.7BQ8_010.0 GBQ8_010details
Ornith 1.5 9B9.7BQ8_010.0 GBQ8_010details
Qwen3.5 9B Base9.7BQ8_010.0 GBQ8_010details
MiMo V2.6 Distill Qwen 9B9.4BQ8_010.0 GBQ8_010details
glm 4 9b9.4BQ6_K13.4 GBno12details
internlm3 8b instruct8.8BQ8_09.7 GBQ8_011details
granite 4.2 8b8.8BQ8_010.5 GBQ8_011details
granite 4.1 8b8.8BQ8_010.5 GBQ8_011details
Qwen3 VL 8B Instruct8.8BQ8_09.9 GBQ8_011details
Molmo2 8B8.7BQ8_09.9 GBQ8_011details
LFM2.5 8B A1B8.5B (1.0B act.)Q8_09.1 GBQ8_094details
Qwen2.5 VL 7B Instruct8.3BQ8_08.7 GBQ8_012details
UI TARS 1.5 7B8.3BQ8_08.7 GBQ8_012details
olmOCR 2 7B 10258.3BQ8_08.7 GBQ8_012details
olmOCR 7B 0225 preview8.3BQ8_08.7 GBQ8_012details
UI TARS 7B DPO8.3BQ8_08.7 GBQ8_012details
Qwen3 8B8.2BQ8_09.9 GBQ8_011details
DeepSeek R1 0528 Qwen3 8B8.2BQ8_09.9 GBQ8_011details
granite 3.3 8b instruct8.2BQ8_09.8 GBQ8_012details
granite 3.1 8b instruct8.2BQ8_09.8 GBQ8_012details
granite 3.2 8b instruct8.2BQ8_09.8 GBQ8_012details
Apertus 8B Instruct 25098.1BQ8_09.6 GBQ8_012details
Llama 3.1 8B Instruct8.0BQ8_09.6 GBQ8_012details
Llama 3.1 8B8.0BQ8_09.6 GBQ8_012details
Hermes 3 Llama 3.1 8B8.0BQ8_09.6 GBQ8_012details
DeepSeek R1 Distill Llama 8B8.0BQ8_09.6 GBQ8_012details
Meta Llama 3 8B Instruct8.0BQ8_09.6 GBQ8_012details
Meta Llama Guard 2 8B8.0BQ8_09.6 GBQ8_012details
Meta Llama 3.1 8B Instruct8.0BQ8_09.6 GBQ8_012details
Ling 3.0 tiny7.9B (1.4B act.)Q8_08.7 GBQ8_068details
MiMo 7B RL7.8BQ8_09.4 GBQ8_012details
internlm2 5 7b chat7.7BQ8_09.2 GBQ8_012details
internlm2 chat 7b7.7BQ8_09.2 GBQ8_012details
Qwen2.5 7B Instruct7.6BQ8_08.7 GBQ8_012details
Qwen2.5 Coder 7B Instruct7.6BQ8_08.7 GBQ8_012details
DeepSeek R1 Distill Qwen 7B7.6BQ8_08.7 GBQ8_012details
Falcon H1R 7B7.6BQ8_08.5 GBQ8_012details
Olmo 3 7B Instruct7.3BQ8_010.3 GBQ8_013details
Olmo 3 1025 7B7.3BQ8_010.3 GBQ8_013details
Olmo 3 7B Think7.3BQ8_010.3 GBQ8_013details
Mistral 7B v0.17.2BQ8_08.6 GBQ8_013details
Mistral 7B Instruct v0.27.2BQ8_08.6 GBQ8_013details
Mistral 7B Instruct v0.17.2BQ8_08.6 GBQ8_013details
falcon 7b7.2BQ8_012.1 GBno13details
falcon 7b instruct7.2BQ8_011.8 GBno14details
granite 4.0 h tiny6.9B (1.5B act.)F1613.6 GBF1634details
OLMoE 1B 7B 09246.9B (1.3B act.)F1614.4 GBQ8_039details
deepseek coder 7b instruct v1.56.9BQ8_011.2 GBno14details
deepseek coder 6.7b instruct6.7BQ8_011.1 GBno14details
deepseek coder 6.7b base6.7BQ8_011.1 GBno14details
Llama 2 7b hf6.7BQ8_011.1 GBno14details
Yi 6B 200K6.1BF1612.3 GBF168.2details
Yi 6B6.1BF1612.3 GBF168.2details
Molmo2 4B4.9BF1610.0 GBF1611details
WeVisOne VL 4B4.8BF1610.0 GBF1611details
Mage VL Cowork4.7BF1610.0 GBF1611details
Qianfan OCR4.7BF1610.1 GBF1611details
Qwen3.5 4B4.7BF169.0 GBF1612details
S5 nano4.7BF169.2 GBF1612details
LightOnOCR 3 4B4.5BF169.0 GBF1612details
SingVL S1 4B4.5BF169.0 GBF1612details
Intern Decision 4B4.5BF169.0 GBF1612details
Qwen3 VL 4B Instruct4.4BF169.3 GBF1612details
gemma 3 4b it4.3BF168.4 GBF1613details
medgemma 1.5 4b it4.3BF168.4 GBF1613details
medgemma 4b it4.3BF168.4 GBF1613details
gemma 3 4b pt4.3BF169.2 GBF1612details
Phi 3.5 vision instruct4.1BF1611.2 GBQ3_K_M12details
Qwen3 4B4.0BF169.3 GBF1612details
Qwen3 4B Instruct 25074.0BF169.3 GBF1612details
granite vision 4.1 4b4.0BF167.6 GBF1615details
Phi 4 mini instruct3.8BF168.9 GBF1613details
Phi 3 mini 4k instruct3.8BF1610.6 GBQ3_K_M13details
Phi 3.5 mini instruct3.8BF1610.6 GBQ3_K_M13details
Phi 3 mini 128k instruct3.8BF1610.6 GBQ3_K_M13details
Qwen2.5 VL 3B Instruct3.8BF166.7 GBF1616details
granite 4.2 3b3.7BF168.0 GBF1614details
granite 4.1 3b llmon preview3.7BF168.0 GBF1614details
Raon 1 nano3.5BF167.0 GBF1616details
granite 4.1 3b3.4BF167.6 GBF1615details
granite 4.0 micro3.4BF167.6 GBF1615details
DeepSeek OCR 23.4B (1.2B act.)F166.6 GBF1648details
deepseek vl2 tiny3.4B (1.2B act.)F167.4 GBF1643details
Unlimited OCR 23.3B (1.1B act.)F166.6 GBF1649details
Unlimited OCR3.3B (1.1B act.)F166.6 GBF1650details
DeepSeek OCR3.3B (1.1B act.)F166.6 GBF1650details
granite 3.1 3b a800m instruct3.3B (883M act.)F167.1 GBF1657details
Llama 3.2 3B Instruct3.2BF1613.5 GBQ8_07.8details
Llama 3.2 3B3.2BF167.5 GBF1616details
Hermes 3 Llama 3.2 3B3.2BF167.5 GBF1616details
d1 3B3.1BF1610.8 GBF169.3details
LFM2.5 VL 3B3.1BF1610.8 GBF169.3details
Qwen2.5 3B Instruct3.1BF166.7 GBF1616details
SmolLM3 3B3.1BF166.9 GBF1616details
SmolLM3 3B Base3.1BF166.9 GBF1616details
phi 22.8BF168.2 GBQ8_018details
LFM2.5 2.6B2.7BF1610.8 GBF169.3details
granite 3.3 2b instruct2.5BF165.8 GBF1620details
MiniCPM5 2B2.5BF165.7 GBF1620details
WeVisOne VL 2B2.4BF165.4 GBF1625details
Qwen3.5 2B2.3BF164.5 GBF1626details
SmolVLM Instruct2.2BF165.4 GBF1628details
Qwen3 1.7B2.0BF164.8 GBF1629details
Qwen3.8 27B DSpine b162.0BF164.8 GBF1625details
d1 2B1.9BF167.2 GBF1614details
LFM3 VL 2B1.9BF167.2 GBF1614details
DeepSeek R1 Distill Qwen 1.5B1.8BF164.2 GBF1628details
LFM3 1.7B1.7BF167.2 GBF1614details
LFM3 1.7B Base1.7BF164.0 GBF1629details
SmolLM2 1.7B Instruct1.7BF165.2 GBF1629details
SmolLM2 1.7B1.7BF165.2 GBF1629details
SmolLM 1.7B1.7BF165.2 GBF1629details
granite 4.0 1b1.6BF164.3 GBF1631details
Qwen2.5 1.5B Instruct1.5BF163.8 GBF1632details
OLMo 2 0425 1B1.5BF164.4 GBF1634details
deepseek coder 1.3b instruct1.3BF164.5 GBF1637details
Llama 3.2 1B1.2BF163.2 GBF1640details
Llama 3.2 1B Instruct1.2BF165.5 GBF1620details
LFM2.5 1.2B Instruct1.2BF165.0 GBF1621details
HunyuanOCR1.1BF163.5 GBF1645details
PaddleOCR VL 1.6959MF162.5 GBF1652details
Qwen3.5 0.8B873MF162.4 GBF1666details
LightOnOCR 3 0.8B853MF162.4 GBF1666details
FastVLM 0.5B759MF162.2 GBF1666details
Qwen3 0.6B752MF162.7 GBF1683details
GOT OCR2 0716MF162.8 GBF1670details
Nemotron 3.5 Parse631MF161.9 GBF1677details
GOT OCR 2.0 hf561MF162.5 GBF1689details
SmolVLM2 500M Video Instruct507MF161.6 GBF16122details
SmolVLM 500M Instruct507MF161.6 GBF16122details
Qwen2.5 0.5B Instruct494MF161.7 GBF16101details
LFM2.5 VL 450M449MF161.9 GBF1670details
SmolLM2 360M Instruct362MF161.5 GBF16138details
SmolLM2 360M362MF161.5 GBF16138details
LFM2.5 350M354MF161.9 GBF1670details
LFM2 350M354MF161.4 GBF16141details

Just too big at 4-bit

These need up to 35% more memory than the Mac M3 24GB has at Q4_K_M and 8K context. A smaller quant or shorter context may work.

ModelQ4_K_M needsShort by
GLiDE 27B16.2 GB0.1 GB
MiniMax H3 Image PE16.8 GB0.7 GB
Qwen3.8 27B16.8 GB0.7 GB
Qwen3.5 27B17.0 GB0.9 GB
Qwen3.6 27B17.0 GB1.0 GB
diffusiongemma 26B A4B it17.1 GB1.0 GB
gemma 4 26B A4B it17.2 GB1.1 GB
gemma 3 27b it17.3 GB1.3 GB
Arete 27B17.4 GB1.3 GB
Muse Glimmer 30B17.9 GB1.8 GB
GLM 4.7 Flash18.2 GB2.1 GB
Hy Memory 30B A3B18.5 GB2.4 GB
Qwen3 Coder 30B A3B Instruct18.7 GB2.6 GB
granite 4.0 h small18.9 GB2.8 GB
granite 4.2 30b19.1 GB3.0 GB

Questions

What is the biggest model that runs on the Mac M3 24GB?

Of the models we track, the largest that fits at 8K context is flint (40B) at Q2_K, needing about 14.8 GB.

How much of the Mac M3 24GB's memory can a model use?

About 16 GB of 24 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.