Apple · apple

Apple M4 Max

Apple M4 Max has 48 GB of unified memory at 546 GB/s — about 33.48 GiB usable after driver and compositor overhead. 2023 of 2118 indexed models fit at 64K context with q4_0 KV. Note only 36 GB of its 48 GB is allocatable to the GPU.

Spec sheet· bandwidth, theoreticalFrom the file· fit from summed bytesPredicted· speed
Memory
48 GB
LPDDR5X-8533
Bandwidth
546 GB/s
512-bit bus
Tensor FP16
—
dense
TDP
—
Context4K8K16K32K64K128K
KV cachef16q8_0q4_0quantizing the KV cache is a ~2× lever on the dominant term at long context
text 1737vision language 182video 16audio tts 21image 2embedding 26audio asr 39

What fits at 64K context

largest quantization that fits, per model · 2023 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
CalmeRys-78B-Orpo-v0.1I1-IQ2_M78.0B29.27 GiB6.05 GiB36.00 GiB0.00 GiB12±8.3%
calme-2.3-rys-78bIQ2_M78.0B29.27 GiB6.05 GiB36.00 GiB0.00 GiB12±8.3%
Delphi-25B-SimpleRL-MathI1-Q5_K_M25.0B16.52 GiB18.83 GiB35.98 GiB0.02 GiB12±8.3%
Qwen2.5-VL-72B-InstructUD-IQ3_XXS73.4B29.67 GiB5.63 GiB35.98 GiB0.02 GiB12±8.3%
Kimi-Dev-72BUD-IQ3_XXS72.7B29.67 GiB5.63 GiB35.97 GiB0.03 GiB12±8.3%
Mistral-Small-4-119B-2603MoEUD-IQ2_M119B34.99 GiB0.40 GiB35.97 GiB0.03 GiB52±37%
Rombo-LLM-V3.0-Qwen-72bI1-IQ3_XXS72.7B29.66 GiB5.63 GiB35.96 GiB0.04 GiB12±8.3%
Qwen2.5-72B-Instruct-abliteratedI1-IQ3_XXS72.7B29.66 GiB5.63 GiB35.96 GiB0.04 GiB12±8.3%
Qwen2.5-72B-Instruct-abliterated-v2I1-IQ3_XXS72.7B29.66 GiB5.63 GiB35.96 GiB0.04 GiB12±8.3%
HuatuoGPT-o1-72BIQ3_XXS72.7B29.66 GiB5.63 GiB35.96 GiB0.04 GiB12±8.3%
MiroThinker-v1.0-72BI1-IQ3_XXS72.7B29.66 GiB5.63 GiB35.96 GiB0.04 GiB12±8.3%
EVA-Qwen2.5-72B-v0.2IQ3_XXS72.7B29.66 GiB5.63 GiB35.96 GiB0.04 GiB12±8.3%
Qwen2.5-Math-72B-InstructIQ3_XXS72.7B29.66 GiB5.63 GiB35.96 GiB0.04 GiB12±8.3%
Qwen2.5-72B-InstructIQ3_XXS72.7B29.66 GiB5.63 GiB35.96 GiB0.04 GiB12±8.3%
Malaysian-Qwen2.5-72B-InstructI1-IQ3_XXS72.7B29.66 GiB5.63 GiB35.96 GiB0.04 GiB12±8.3%
Qwen2.5-72BI1-IQ3_XXS72.7B29.66 GiB5.63 GiB35.96 GiB0.04 GiB12±8.3%
magnum-v4-72bI1-IQ3_XXS72.7B29.66 GiB5.63 GiB35.96 GiB0.04 GiB12±8.3%
KAT-Dev-72B-ExpIQ3_XXS72.7B29.66 GiB5.63 GiB35.96 GiB0.04 GiB12±8.3%
Homer-v1.0-Qwen2.5-72BIQ3_XXS72.7B29.66 GiB5.63 GiB35.96 GiB0.04 GiB12±8.3%
Tower-Plus-72B-ultra-uncensored-hereticI1-IQ3_XXS72.7B29.66 GiB5.63 GiB35.96 GiB0.04 GiB12±8.3%
Chronos-Platinum-72BIQ3_XXS72.7B29.66 GiB5.63 GiB35.96 GiB0.04 GiB12±8.3%
UI-TARS-72B-DPOIQ3_XXS73.4B29.66 GiB5.63 GiB35.96 GiB0.04 GiB12±8.3%
Llama-4-Scout-17B-16E-InstructMoEKV unresolvedIQ2_S109B31.98 GiB3.38 GiB35.93 GiB0.07 GiB30±37%
Hunyuan-A13B-InstructMoEQ3_K_S80.4B33.10 GiB2.25 GiB35.90 GiB0.10 GiB12±8.3%
Qwen3.6-27B-Fable-Fusion-711-Uncensored-Heretic-NM-DAU-MTPQ4_K_M27.8B34.04 GiB1.13 GiB35.77 GiB0.23 GiB12±8.3%
deepseek-llm-67b-chatI1-IQ3_M67.4B28.43 GiB6.68 GiB35.76 GiB0.24 GiB12±8.3%
deepseek-llm-67b-baseI1-IQ3_M67.4B28.43 GiB6.68 GiB35.76 GiB0.24 GiB12±8.3%
openbuddy-deepseek-67b-v15.3-4kI1-IQ3_M67.4B28.43 GiB6.68 GiB35.76 GiB0.24 GiB12±8.3%
llm-surgery-dark-arts-gpt-oss-60b-96a12MoEI1-Q4_160.9B34.76 GiB0.43 GiB35.73 GiB0.27 GiB33±37%
Qwen3-Coder-NextMoEUD-Q3_K_M79.7B33.47 GiB1.69 GiB35.70 GiB0.30 GiB45±37%
Qwen3.6-27B-uncensored-heretic-v2Q8_027.4B33.96 GiB1.13 GiB35.69 GiB0.31 GiB12±8.3%
Hypernova-60B-2605MoEI1-Q4_158.7B34.48 GiB0.57 GiB35.59 GiB0.41 GiB45±37%
Noromaid-20b-v0.1.1I1-Q5_K_M20.0B13.18 GiB21.80 GiB35.57 GiB0.43 GiB12±8.3%
Nethena-20BQ5_K_M20.0B13.18 GiB21.80 GiB35.57 GiB0.43 GiB12±8.3%
Gemma-4-31B-Isometry-RPQ8_032.7B31.79 GiB3.14 GiB35.56 GiB0.44 GiB12±8.3%
Prosopon-31BQ8_032.7B31.79 GiB3.14 GiB35.56 GiB0.44 GiB12±8.3%
Gemma-4-Novelist-Eclipse-31BQ8_032.7B31.79 GiB3.14 GiB35.56 GiB0.44 GiB12±8.3%
Giftige-Blume-31B-v1-StyleSwapQ8_032.7B31.79 GiB3.14 GiB35.56 GiB0.44 GiB12±8.3%
G4-MeroMero-31B-StyleSwapQ8_032.7B31.79 GiB3.14 GiB35.56 GiB0.44 GiB12±8.3%
Gemma-4-31B-StyleTune-heretic-araQ8_032.7B31.79 GiB3.14 GiB35.56 GiB0.44 GiB12±8.3%
Pantheon-Reasoning-31B-1.1Q8_032.7B31.79 GiB3.14 GiB35.56 GiB0.44 GiB12±8.3%
Gemma-4-31B-StyleTuneQ8_032.7B31.79 GiB3.14 GiB35.56 GiB0.44 GiB12±8.3%
Barcenas-StyleTune-31B-FableQ8_032.1B31.79 GiB3.14 GiB35.56 GiB0.44 GiB12±8.3%
Skyfall-31B-v4.2-hereticQ8_031.4B31.03 GiB3.80 GiB35.50 GiB0.50 GiB13±8.3%
Skyfall-31B-v4.2Q8_031.4B31.03 GiB3.80 GiB35.50 GiB0.50 GiB13±8.3%
Qwen3.6-35B-A3B-REAM-192-hereticMoEQ5_K_M27.0B34.58 GiB0.35 GiB35.49 GiB0.51 GiB49±37%
Gemma-The-Writer-N-Restless-Quill-10B-UncensoredQ8_010.0B31.41 GiB3.46 GiB35.46 GiB0.54 GiB12±8.3%
c4ai-command-r-08-2024Q8_032.3B31.97 GiB2.81 GiB35.45 GiB0.55 GiB13±8.3%
Huihui-GLM-4.7-Flash-abliterated-57BMoEI1-Q4_157.3B32.47 GiB2.35 GiB35.42 GiB0.58 GiB34±37%
GLM-4.5VMoEI1-IQ1_S108B31.56 GiB3.23 GiB35.37 GiB0.63 GiB31±37%
Devstral-2-123B-Instruct-2512UD-IQ1_M125B28.47 GiB6.19 GiB35.36 GiB0.64 GiB13±8.3%
Qwen3-42B-A3B-2507-Thinking-Abliterated-uncensored-TOTAL-RECALL-v2-Medium-MASTER-CODERMoEI1-Q6_K42.4B32.43 GiB2.36 GiB35.33 GiB0.67 GiB34±37%
Darwin-35B-A3B-OpusMoEQ8_036.0B34.38 GiB0.35 GiB35.29 GiB0.71 GiB53±37%
Aurora-Code-1MoEQ8_034.7B34.38 GiB0.35 GiB35.29 GiB0.71 GiB53±37%
grug-35b-v2MoEQ8_035.1B34.38 GiB0.35 GiB35.29 GiB0.71 GiB53±37%
grug-35bMoEQ8_035.1B34.38 GiB0.35 GiB35.29 GiB0.71 GiB53±37%
WorldSim-Opus-3.6-35B-A3BMoEQ8_035.1B34.38 GiB0.35 GiB35.29 GiB0.71 GiB53±37%
Qwen3.6-35B-A3B-AnkoMoEQ8_035.1B34.38 GiB0.35 GiB35.29 GiB0.71 GiB53±37%
KAT-Coder-V2.5-DevMoEQ8_034.7B34.38 GiB0.35 GiB35.29 GiB0.71 GiB53±37%
Ornith-1.0-35BMoEQ8_034.7B34.38 GiB0.35 GiB35.29 GiB0.71 GiB53±37%
●From the file◐Predictedwhat these mean

Speed is modeled, not measured: decode is memory-bandwidth bound, so tokens per second is bytes read per token against achievable bandwidth. Mixture-of-experts models carry a wider band because only the routed experts are read each step, and few have been measured publicly.

Questions people ask

What AI models can a Apple M4 Max run?
2023 of 2118 indexed open-weight models fit a Apple M4 Max at 65,536 context with q4_0 KV cache, the largest being CalmeRys-78B-Orpo-v0.1 at I1-IQ2_M. That covers text, vision-language, image, video and speech models.
How much usable memory does a Apple M4 Max actually have?
Its nameplate is 48 GB, but about 33.48 GiB is available to a model once driver and compositor overhead is accounted for, and only 36 GB of the pool can be allocated to the GPU at all.
Is a Apple M4 Max fast for local AI?
Its memory bandwidth is 546 GB/s, and that figure — not teraflops — is what governs token generation speed. Capacity decides what you can run; bandwidth decides how fast it runs.