Apple · apple

Apple M2 Max

Apple M2 Max has 64 GB of unified memory at 410 GB/s — about 44.64 GiB usable after driver and compositor overhead. 2029 of 2118 indexed models fit at 128K context with q4_0 KV. Note only 48 GB of its 64 GB is allocatable to the GPU.

Spec sheet· bandwidth, theoreticalFrom the file· fit from summed bytesPredicted· speed
Memory
64 GB
LPDDR5-6400
Bandwidth
410 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 1741vision language 184image 2audio tts 21video 16audio asr 39embedding 26

What fits at 128K context

largest quantization that fits, per model · 2029 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
Huihui-Qwen3-Coder-Next-abliteratedMoEI1-Q4_179.7B46.60 GiB0.84 GiB47.98 GiB0.02 GiB35±37%
Huihui-GLM-4.7-Flash-abliterated-57BMoEI1-Q6_K57.3B42.64 GiB4.71 GiB47.94 GiB0.06 GiB18±37%
Ornith-1.0-35B-AEON-Ultimate-Uncensored-BF16MoEQ5_K_M35.1B46.64 GiB0.70 GiB47.89 GiB0.11 GiB33±37%
Qwen3.5-88BMoEI1-Q4_087.7B46.41 GiB0.84 GiB47.84 GiB0.16 GiB30±37%
Devstral-2-123B-Instruct-2512IQ2_XS125B34.75 GiB12.38 GiB47.83 GiB0.17 GiB7±8.3%
Mistral-Medium-3.5-128BI1-IQ2_XS128B34.75 GiB12.38 GiB47.83 GiB0.17 GiB7±8.3%
XORTRON-NXTXPRTXXLI1-IQ2_XS128B34.75 GiB12.38 GiB47.83 GiB0.17 GiB7±8.3%
Mistral-Small-4-119B-2603MoEIQ3_XXS119B46.44 GiB0.79 GiB47.82 GiB0.18 GiB33±37%
Hunyuan-A13B-InstructMoEIQ4_NL80.4B42.77 GiB4.50 GiB47.81 GiB0.19 GiB7±8.3%
GPT-NeoX-20B-ErebusI1-Q3_K_M20.6B10.03 GiB37.13 GiB47.80 GiB0.20 GiB7±8.3%
GLM-4.5VMoEI1-Q2_K_S108B40.74 GiB6.47 GiB47.79 GiB0.21 GiB16±37%
Step-3.5-Flash-REAP-121B-A11BI1-IQ2_S121B33.40 GiB13.79 GiB47.76 GiB0.24 GiB7±8.3%
deepseek-llm-67b-chatI1-IQ4_XS67.4B33.71 GiB13.36 GiB47.72 GiB0.28 GiB7±8.3%
deepseek-llm-67b-baseI1-IQ4_XS67.4B33.71 GiB13.36 GiB47.72 GiB0.28 GiB7±8.3%
openbuddy-deepseek-67b-v15.3-4kI1-IQ4_XS67.4B33.71 GiB13.36 GiB47.72 GiB0.28 GiB7±8.3%
command-a-plus-05-2026-bf16MoEIQ1_S219B45.87 GiB1.24 GiB47.67 GiB0.33 GiB25±37%
HunyuanImage-2.1Q5_017.5B47.04 GiB0.00 GiB47.64 GiB0.36 GiB7±8.3%
Qwen3-Coder-NextMoEQ4_K_S79.7B43.71 GiB3.38 GiB47.62 GiB0.38 GiB25±37%
Qwen3-Next-80B-A3B-ThinkingMoEQ4_K_S81.3B43.71 GiB3.38 GiB47.62 GiB0.38 GiB25±37%
Qwen3-Next-80B-A3B-InstructMoEQ4_K_S81.3B43.71 GiB3.38 GiB47.62 GiB0.38 GiB25±37%
Qwen3-Coder-Next-REAMMoEI1-Q6_K60.3B46.22 GiB0.84 GiB47.60 GiB0.40 GiB34±37%
c4ai-command-r-plus-08-2024IQ3_XXS104B37.87 GiB9.00 GiB47.60 GiB0.40 GiB7±8.3%
Maenad-70BIQ4_XS70.6B35.64 GiB11.25 GiB47.57 GiB0.43 GiB7±8.3%
DeepSeek-R1-Distill-Llama-70B-Uncensored-v2-Unbiased-ReasonerIQ4_XS70.6B35.64 GiB11.25 GiB47.57 GiB0.43 GiB7±8.3%
Strawberrylemonade-L3-70B-v1.2IQ4_XS70.6B35.64 GiB11.25 GiB47.57 GiB0.43 GiB7±8.3%
Llama-3.3_70_b_uncensored_continuedIQ4_XS70.6B35.64 GiB11.25 GiB47.57 GiB0.43 GiB7±8.3%
Llama-3.3-70B-Instruct-abliteratedIQ4_XS70.6B35.64 GiB11.25 GiB47.57 GiB0.43 GiB7±8.3%
grok-oss-Revenant-70BIQ4_XS70.6B35.64 GiB11.25 GiB47.57 GiB0.43 GiB7±8.3%
Hermes-4-70B-hereticIQ4_XS70.6B35.64 GiB11.25 GiB47.57 GiB0.43 GiB7±8.3%
Golem-70B-v1bIQ4_XS70.6B35.64 GiB11.25 GiB47.57 GiB0.43 GiB7±8.3%
DeepSeek-R1-Distill-Llama-70B-hereticIQ4_XS70.6B35.64 GiB11.25 GiB47.57 GiB0.43 GiB7±8.3%
Infinity-Instruct-7M-Gen-Llama3_1-70BIQ4_XS70.6B35.64 GiB11.25 GiB47.57 GiB0.43 GiB7±8.3%
ALIA-40b-fc-2606Q8_040.4B40.02 GiB6.75 GiB47.43 GiB0.57 GiB7±8.3%
ALIA-40b-instruct-2606Q8_040.4B40.02 GiB6.75 GiB47.43 GiB0.57 GiB7±8.3%
Laguna-S-2.1MoEUD-IQ3_S118B45.10 GiB1.73 GiB47.40 GiB0.60 GiB28±37%
Llama-4-Scout-17B-16E-InstructMoEKV unresolvedQ2_K109B40.03 GiB6.75 GiB47.36 GiB0.64 GiB16±37%
Apertus-70B-Instruct-2509IQ4_XS70.6B35.36 GiB11.25 GiB47.34 GiB0.66 GiB7±8.3%
MiniMax-M2.1-REAP-139B-A10BMoEI1-IQ2_XS139B38.06 GiB8.72 GiB47.31 GiB0.69 GiB14±37%
m51Lab-MiniMax-M2.7-REAP-139B-A10BMoEI1-IQ2_XS139B38.06 GiB8.72 GiB47.31 GiB0.69 GiB14±37%
Delphi-25B-SimpleRL-MathI1-IQ3_XXS25.0B9.03 GiB37.65 GiB47.31 GiB0.69 GiB7±8.3%
GLM-4.6VMoEIQ2_M108B40.26 GiB6.47 GiB47.30 GiB0.70 GiB16±37%
Hermes-4-70BIQ4_XS70.6B35.33 GiB11.25 GiB47.26 GiB0.74 GiB7±8.3%
Llama-3.3-70B-InstructIQ4_XS70.6B35.33 GiB11.25 GiB47.26 GiB0.74 GiB7±8.3%
DeepSeek-R1-Distill-Llama-70BIQ4_XS70.6B35.33 GiB11.25 GiB47.26 GiB0.74 GiB7±8.3%
Qwen3-42B-A3B-2507-Thinking-Abliterated-uncensored-TOTAL-RECALL-v2-Medium-MASTER-CODERMoEQ8_042.4B41.98 GiB4.71 GiB47.24 GiB0.76 GiB18±37%
calme-2.4-llama3-70bIQ4_XS70.6B35.30 GiB11.25 GiB47.23 GiB0.77 GiB7±8.3%
calme-2.2-llama3-70bIQ4_XS70.6B35.30 GiB11.25 GiB47.23 GiB0.77 GiB7±8.3%
Rombos-LLM-70b-Llama-3.3I1-IQ4_XS70.6B35.30 GiB11.25 GiB47.23 GiB0.77 GiB7±8.3%
L3.3-Electra-R1-70bI1-IQ4_XS70.6B35.30 GiB11.25 GiB47.23 GiB0.77 GiB7±8.3%
L3.3-70B-Magnum-v4-SEIQ4_XS70.6B35.30 GiB11.25 GiB47.23 GiB0.77 GiB7±8.3%
Latxa-Llama-3.1-70B-Instruct-v2I1-IQ4_XS70.6B35.30 GiB11.25 GiB47.23 GiB0.77 GiB7±8.3%
Llama-3.1-Nemotron-70B-Instruct-HFI1-IQ4_XS70.6B35.30 GiB11.25 GiB47.23 GiB0.77 GiB7±8.3%
L3.3-70B-Euryale-v2.3I1-IQ4_XS70.6B35.30 GiB11.25 GiB47.23 GiB0.77 GiB7±8.3%
Hermes-3-Llama-3.1-70BIQ4_XS70.6B35.30 GiB11.25 GiB47.23 GiB0.77 GiB7±8.3%
Llama-3.1-70BIQ4_XS70.6B35.30 GiB11.25 GiB47.23 GiB0.77 GiB7±8.3%
Anubis-70B-v1.2IQ4_XS70.6B35.30 GiB11.25 GiB47.23 GiB0.77 GiB7±8.3%
DeepSeek-R1-Distill-Llama-70B-abliteratedI1-IQ4_XS70.6B35.30 GiB11.25 GiB47.23 GiB0.77 GiB7±8.3%
llama-3-firefunction-v2IQ4_XS70.6B35.30 GiB11.25 GiB47.23 GiB0.77 GiB7±8.3%
Legion-V2.1-LLaMa-70BI1-IQ4_XS70.6B35.30 GiB11.25 GiB47.23 GiB0.77 GiB7±8.3%
Assistant_Pepe_70BI1-IQ4_XS70.6B35.30 GiB11.25 GiB47.23 GiB0.77 GiB7±8.3%
●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 M2 Max run?
2029 of 2118 indexed open-weight models fit a Apple M2 Max at 131,072 context with q4_0 KV cache, the largest being Huihui-Qwen3-Coder-Next-abliterated at I1-Q4_1. That covers text, vision-language, image, video and speech models.
How much usable memory does a Apple M2 Max actually have?
Its nameplate is 64 GB, but about 44.64 GiB is available to a model once driver and compositor overhead is accounted for, and only 48 GB of the pool can be allocated to the GPU at all.
Is a Apple M2 Max fast for local AI?
Its memory bandwidth is 410 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.