RunsOnMy ModelsGPUs & MacsAPI pricesVRAM calculatorNew

What AI models run on the Mac M2 Max 96GB?

The Mac M2 Max 96GB has 96 GB of unified memory (400 GB/s). About 72.0 GB is usable for a model. 219 of 285 open models we track fit at 8K context.

219models fit (8K ctx)
218still fit at 32K ctx
199Blargest that fits
400 GB/smemory bandwidth

Models that fit, largest first

ModelParamsBest quant (8K)Needs32K contextMax tok/s (est.)
Step 3.5 Flash199BQ2_K69.0 GBno5.8details
Qwen3.8 Flash Next180B (62B act.)Q2_K63.9 GBQ2_K17details
Ling 3.0 flash Sante127B (5.6B act.)Q3_K_M58.3 GBQ3_K_M148details
Ling 3.0 flash127B (5.6B act.)Q3_K_M56.0 GBQ3_K_M154details
Ornith 1.5 122B125B (13B act.)Q3_K_M58.0 GBQ3_K_M64details
gpt oss 120b117B (5.8B act.)F1661.9 GBF16124details
gpt oss safeguard 120b117B (5.8B act.)Q8_060.1 GBQ8_0128details
GLM 4.5 Air110B (15B act.)Q3_K_M55.4 GBQ3_K_M52details
Llama 4 Scout 17B 16E Instruct109B (18B act.)Q4_K_M63.2 GBQ3_K_M37details
Llama 4 Scout 17B 16E109B (18B act.)Q4_K_M64.1 GBQ3_K_M36details
GLM 4.5V108B (14B act.)Q4_K_M61.4 GBQ3_K_M47details
GLM 4.6V108B (14B act.)Q3_K_M53.8 GBQ3_K_M54details
Hunyuan A13B Instruct80B (13B act.)Q6_K63.0 GBQ5_K_M38details
Kimi Dev 72B73BQ6_K63.1 GBQ5_K_M6.2details
Llama 3.3 70B Instruct71BQ6_K57.0 GBQ6_K6.9details
DeepSeek R1 Distill Llama 70B71BQ6_K57.0 GBQ6_K6.9details
Llama 3.1 70B71BQ6_K57.1 GBQ6_K6.9details
Kimi Linear 48B A3B Instruct49B (3.0B act.)Q8_049.6 GBQ8_0125details
Nous Hermes 2 Mixtral 8x7B DPO47B (13B act.)Q8_047.7 GBQ8_029details
falcon 40b42BQ8_042.9 GBQ8_09.0details
flint40B (1.3B act.)Q8_040.3 GBQ8_0301details
Maxwell V1 Flash37B (2.3B act.)Q8_037.6 GBQ8_0166details
Hermes 4.3 36B36BQ8_038.5 GBQ8_010details
Seed OSS 36B Instruct36BQ8_038.5 GBQ8_010details
Qwen3.6 35B A3B36B (3.0B act.)Q8_036.2 GBQ8_0127details
Ornith 1.5 35B A3B36B (3.0B act.)Q8_036.2 GBQ8_0127details
Qwen3.5 35B A3B36B (3.0B act.)Q8_035.4 GBQ8_0130details
Axion 1.0 35B35B (3.9B act.)Q8_035.8 GBQ8_096details
Qwen AgentWorld 35B A3B35B (3.0B act.)Q8_035.4 GBQ8_0125details
Yi 34B 200K34BQ8_036.4 GBQ8_011details
Yi 34B34BQ8_036.4 GBQ8_011details
Yi 34B Chat34BQ8_036.4 GBQ8_011details
Yi 1.5 34B Chat34BQ8_036.4 GBQ8_011details
Nous Hermes 2 Yi 34B34BQ8_036.4 GBQ8_011details
DeepSeek R1 Distill Qwen 32B33BF1663.7 GBQ8_06.1details
Qwen3 32B33BF1663.7 GBQ8_06.1details
GLM Z1 32B 041433BF1661.8 GBF166.1details
GLM 4 32B 041433BF1661.8 GBF166.1details
granite 4.0 h small32B (8.8B act.)F1660.7 GBF1623details
gemma 4 31B it31BF1660.1 GBF166.5details
GLM 4.7 Flash31B (4.6B act.)F1656.9 GBF1645details
Qwen3 Coder 30B A3B Instruct31B (3.0B act.)F1658.3 GBF1667details
North Mini Code 1.030B (3.3B act.)F1658.2 GBF1660details
Hy Memory 30B A3B30B (3.0B act.)F1657.4 GBF1667details
Muse Glimmer 30B30BF1652.9 GBF167.2details
granite 4.2 30b29BF1657.1 GBF166.8details
Qwen3.8 27B28BF1652.3 GBF167.3details
Qwen3.6 27B28BF1651.5 GBF167.4details
Qwen3.5 27B28BF1651.5 GBF167.4details
gemma 3 27b it27BF1652.2 GBF167.4details
Arete 27B27BF1651.5 GBF167.4details
MiniMax H3 Image PE27BF1652.3 GBF167.3details
Trinity Mini26B (2.0B act.)F1649.7 GBF16102details
GLiDE 27B26BF1650.0 GBF167.7details
diffusiongemma 26B A4B it26B (4.0B act.)F1648.5 GBF1651details
gemma 4 26B A4B it26B (4.0B act.)F1648.4 GBF1651details
LFM2.5 24B A2B24B (2.0B act.)Q8_024.5 GBQ8_0188details
ERNIE 4.5 21B A3B PT22BF1641.7 GBF169.2details
ERNIE 4.5 21B A3B Thinking22BF1641.7 GBF169.2details
gpt oss safeguard 20b22B (4.8B act.)F1613.8 GBF16130details
gpt oss 20b21B (4.2B act.)F1613.8 GBF16145details
Groot 19B A3B 26102319B (3.0B act.)F1636.4 GBF1668details
Kimi VL A3B Thinking 250616B (3.0B act.)F1660.4 GBF1634details
Kimi VL A3B Instruct16B (3.0B act.)F1631.5 GBF1667details
Kimi VL A3B Thinking16B (3.0B act.)F1631.5 GBF1667details
deepseek moe 16b base16B (2.8B act.)F1632.8 GBF1671details
LLaDA2.1 mini16B (1.4B act.)F1631.3 GBF16140details
Moonlight 16B A3B Instruct16B (3.0B act.)F1630.7 GBF1667details
DeepSeek Coder V2 Lite Instruct16B (2.7B act.)F1630.1 GBF1675details
DeepSeek V2 Lite16B (2.7B act.)F1630.1 GBF1675details
DeepSeek V2 Lite Chat16B (2.7B act.)F1630.1 GBF1675details
Qwen2.5 Coder 14B Instruct15BF1629.7 GBF1614details
DeepSeek R1 Distill Qwen 14B15BF1629.7 GBF1614details
Qwen3 14B15BF1629.4 GBF1614details
phi 415BF1629.5 GBF1614details
gemma 3 12b it12BF1623.6 GBF1617details
Llama Guard 4 12B12BF1624.6 GBF1617details
Llama 3.2 11B Vision Instruct11BF1620.1 GBF1620details
Llama 3.2 11B Vision11BF1621.7 GBF1619details
GLM 4.6V Flash10BF1618.5 GBF1621details
GLM 4.1V 9B Thinking10BF1618.5 GBF1621details
Step3 VL 10B10BF1632.3 GBF1612details
SenseNova Loop 9B Preview9.7BF1617.8 GBF1622details
Qwen3.5 9B9.7BF1617.8 GBF1622details
Ornith 1.5 9B9.7BF1617.8 GBF1622details
Qwen3.5 9B Base9.7BF1617.8 GBF1622details
MiMo V2.6 Distill Qwen 9B9.4BF1617.8 GBF1622details
glm 4 9b9.4BF1623.2 GBF1621details
internlm3 8b instruct8.8BF1617.4 GBF1623details
granite 4.2 8b8.8BF1618.2 GBF1623details
granite 4.1 8b8.8BF1618.2 GBF1623details
Qwen3 VL 8B Instruct8.8BF1617.1 GBF1624details
Molmo2 8B8.7BF1617.1 GBF1624details
LFM2.5 8B A1B8.5B (1.0B act.)F1632.3 GBF16100details
Qwen2.5 VL 7B Instruct8.3BF1615.3 GBF1626details
UI TARS 1.5 7B8.3BF1615.3 GBF1626details
olmOCR 2 7B 10258.3BF1615.3 GBF1626details
olmOCR 7B 0225 preview8.3BF1616.6 GBF1624details
UI TARS 7B DPO8.3BF1615.3 GBF1626details
Qwen3 8B8.2BF1617.1 GBF1624details
DeepSeek R1 0528 Qwen3 8B8.2BF1617.1 GBF1624details
granite 3.3 8b instruct8.2BF1617.0 GBF1624details
granite 3.1 8b instruct8.2BF1617.0 GBF1624details
granite 3.2 8b instruct8.2BF1617.0 GBF1624details
Apertus 8B Instruct 25098.1BF1616.7 GBF1625details
Llama 3.1 8B Instruct8.0BF1616.6 GBF1625details
Llama 3.1 8B8.0BF1616.6 GBF1625details
Hermes 3 Llama 3.1 8B8.0BF1616.6 GBF1625details
DeepSeek R1 Distill Llama 8B8.0BF1631.6 GBF1612details
Meta Llama 3 8B Instruct8.0BF1616.6 GBF1625details
Meta Llama Guard 2 8B8.0BF1616.6 GBF1625details
Meta Llama 3.1 8B Instruct8.0BF1616.6 GBF1625details
Ling 3.0 tiny7.9B (1.4B act.)F1615.6 GBF16145details
MiMo 7B RL7.8BF1616.4 GBF1626details
internlm2 5 7b chat7.7BF1616.0 GBF1626details
internlm2 chat 7b7.7BF1616.0 GBF1626details
Qwen2.5 7B Instruct7.6BF1615.3 GBF1626details
Qwen2.5 Coder 7B Instruct7.6BF1615.3 GBF1626details
DeepSeek R1 Distill Qwen 7B7.6BF1615.3 GBF1626details
Falcon H1R 7B7.6BF1615.1 GBF1626details
Olmo 3 7B Instruct7.3BF1616.7 GBF1627details
Olmo 3 1025 7B7.3BF1616.7 GBF1627details
Olmo 3 7B Think7.3BF1616.7 GBF1627details
Mistral 7B v0.17.2BF1614.9 GBF1628details
Mistral 7B Instruct v0.27.2BF1614.9 GBF1628details
Mistral 7B Instruct v0.17.2BF1614.9 GBF1628details
falcon 7b7.2BF1618.4 GBF1628details
falcon 7b instruct7.2BF1618.4 GBF1628details
granite 4.0 h tiny6.9B (1.5B act.)F1613.6 GBF16136details
OLMoE 1B 7B 09246.9B (1.3B act.)F1614.4 GBF16156details
deepseek coder 7b instruct v1.56.9BF1617.2 GBF1629details
deepseek coder 6.7b instruct6.7BF1617.0 GBF1630details
deepseek coder 6.7b base6.7BF1617.0 GBF1630details
Llama 2 7b hf6.7BF1617.0 GBF1630details
Yi 6B 200K6.1BF1612.3 GBF1633details
Yi 6B6.1BF1612.3 GBF1633details
Molmo2 4B4.9BF1610.0 GBF1645details
WeVisOne VL 4B4.8BF1610.0 GBF1645details
Mage VL Cowork4.7BF1610.0 GBF1645details
Qianfan OCR4.7BF1610.1 GBF1645details
Qwen3.5 4B4.7BF169.0 GBF1647details
S5 nano4.7BF169.2 GBF1646details
LightOnOCR 3 4B4.5BF169.0 GBF1647details
SingVL S1 4B4.5BF169.0 GBF1647details
Intern Decision 4B4.5BF169.0 GBF1647details
Qwen3 VL 4B Instruct4.4BF169.3 GBF1650details
gemma 3 4b it4.3BF168.4 GBF1651details
medgemma 1.5 4b it4.3BF168.4 GBF1651details
medgemma 4b it4.3BF168.4 GBF1651details
gemma 3 4b pt4.3BF169.2 GBF1647details
Phi 3.5 vision instruct4.1BF1611.2 GBF1648details
Qwen3 4B4.0BF169.3 GBF1650details
Qwen3 4B Instruct 25074.0BF169.3 GBF1650details
granite vision 4.1 4b4.0BF167.6 GBF1659details
Phi 4 mini instruct3.8BF168.9 GBF1652details
Phi 3 mini 4k instruct3.8BF1610.6 GBF1652details
Phi 3.5 mini instruct3.8BF1610.6 GBF1652details
Phi 3 mini 128k instruct3.8BF1610.6 GBF1652details
Qwen2.5 VL 3B Instruct3.8BF166.7 GBF1665details
granite 4.2 3b3.7BF168.0 GBF1655details
granite 4.1 3b llmon preview3.7BF168.0 GBF1655details
Raon 1 nano3.5BF167.0 GBF1664details
granite 4.1 3b3.4BF167.6 GBF1659details
granite 4.0 micro3.4BF167.6 GBF1659details
DeepSeek OCR 23.4B (1.2B act.)F166.6 GBF16193details
deepseek vl2 tiny3.4B (1.2B act.)F167.4 GBF16170details
Unlimited OCR 23.3B (1.1B act.)F166.6 GBF16198details
Unlimited OCR3.3B (1.1B act.)F166.6 GBF16199details
DeepSeek OCR3.3B (1.1B act.)F166.6 GBF16199details
granite 3.1 3b a800m instruct3.3B (883M act.)F167.1 GBF16226details
Llama 3.2 3B Instruct3.2BF1613.5 GBF1631details
Llama 3.2 3B3.2BF167.5 GBF1662details
Hermes 3 Llama 3.2 3B3.2BF167.5 GBF1662details
d1 3B3.1BF1610.8 GBF1637details
LFM2.5 VL 3B3.1BF1610.8 GBF1637details
Qwen2.5 3B Instruct3.1BF166.7 GBF1665details
SmolLM3 3B3.1BF166.9 GBF1665details
SmolLM3 3B Base3.1BF166.9 GBF1665details
phi 22.8BF168.2 GBF1672details
LFM2.5 2.6B2.7BF1610.8 GBF1637details
granite 3.3 2b instruct2.5BF165.8 GBF1679details
MiniCPM5 2B2.5BF165.7 GBF1679details
WeVisOne VL 2B2.4BF165.4 GBF1698details
Qwen3.5 2B2.3BF164.5 GBF16106details
SmolVLM Instruct2.2BF165.4 GBF16110details
Qwen3 1.7B2.0BF164.8 GBF16116details
Qwen3.8 27B DSpine b162.0BF164.8 GBF1698details
d1 2B1.9BF167.2 GBF1658details
LFM3 VL 2B1.9BF167.2 GBF1658details
DeepSeek R1 Distill Qwen 1.5B1.8BF164.2 GBF16112details
LFM3 1.7B1.7BF167.2 GBF1658details
LFM3 1.7B Base1.7BF164.0 GBF16116details
SmolLM2 1.7B Instruct1.7BF165.2 GBF16117details
SmolLM2 1.7B1.7BF165.2 GBF16117details
SmolLM 1.7B1.7BF165.2 GBF16117details
granite 4.0 1b1.6BF164.3 GBF16122details
Qwen2.5 1.5B Instruct1.5BF163.8 GBF16130details
OLMo 2 0425 1B1.5BF164.4 GBF16135details
deepseek coder 1.3b instruct1.3BF164.5 GBF16149details
Llama 3.2 1B1.2BF163.2 GBF16162details
Llama 3.2 1B Instruct1.2BF165.5 GBF1681details
LFM2.5 1.2B Instruct1.2BF165.0 GBF1685details
HunyuanOCR1.1BF163.5 GBF16179details
PaddleOCR VL 1.6959MF162.5 GBF16209details
Qwen3.5 0.8B873MF162.4 GBF16264details
LightOnOCR 3 0.8B853MF162.4 GBF16264details
FastVLM 0.5B759MF162.2 GBF16264details
Qwen3 0.6B752MF162.7 GBF16334details
GOT OCR2 0716MF162.8 GBF16279details
Nemotron 3.5 Parse631MF161.9 GBF16306details
GOT OCR 2.0 hf561MF162.5 GBF16357details
SmolVLM2 500M Video Instruct507MF161.6 GBF16488details
SmolVLM 500M Instruct507MF161.6 GBF16488details
Qwen2.5 0.5B Instruct494MF161.7 GBF16405details
LFM2.5 VL 450M449MF161.9 GBF16281details
SmolLM2 360M Instruct362MF161.5 GBF16551details
SmolLM2 360M362MF161.5 GBF16551details
LFM2.5 350M354MF161.9 GBF16281details
LFM2 350M354MF161.4 GBF16564details

Just too big at 4-bit

These need up to 35% more memory than the Mac M2 Max 96GB has at Q4_K_M and 8K context. A smaller quant or shorter context may work.

ModelQ4_K_M needsShort by
Ornith 1.5 122B72.3 GB0.3 GB
Ling 3.0 flash73.5 GB1.5 GB
Ling 3.0 flash Sante75.9 GB3.9 GB

Questions

What is the biggest model that runs on the Mac M2 Max 96GB?

Of the models we track, the largest that fits at 8K context is Step 3.5 Flash (199B) at Q2_K, needing about 69.0 GB.

How much of the Mac M2 Max 96GB's memory can a model use?

About 72 GB of 96 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.