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. 2051 of 2118 indexed models fit at 4K 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 1762vision language 185image 2audio asr 39audio tts 21video 16embedding 26

What fits at 4K context

largest quantization that fits, per model · 2051 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
Hunyuan-A13B-InstructMoEQ4_180.4B47.26 GiB0.14 GiB47.95 GiB0.05 GiB7±8.3%
Qwen2.5-72B-InstructQ5_072.7B46.88 GiB0.35 GiB47.91 GiB0.09 GiB7±8.3%
Qwen2.5-72BQ5_072.7B46.86 GiB0.35 GiB47.89 GiB0.11 GiB7±8.3%
Behemoth-X-123B-v2IQ3_XS123B46.70 GiB0.39 GiB47.79 GiB0.21 GiB7±8.3%
diffusiongemma-26B-A4B-it-HERETIC-UncensoredMoEBF1625.8B47.07 GiB0.13 GiB47.73 GiB0.27 GiB7±8.3%
diffusiongemma-26B-A4B-itMoEBF1625.8B47.07 GiB0.13 GiB47.73 GiB0.27 GiB7±8.3%
gemma-4-26B-A4B-it-Claude-Opus-DistillMoEBF1626.5B47.04 GiB0.13 GiB47.70 GiB0.30 GiB7±8.3%
gemma-4-26B-A4B-it-Claude-Opus-Distill-v2MoEBF1626.5B47.04 GiB0.13 GiB47.70 GiB0.30 GiB7±8.3%
G4-MeroMero-26B-A4BMoEBF1625.8B47.04 GiB0.13 GiB47.70 GiB0.30 GiB7±8.3%
gemma-4-26B-A4B-it-qat-q4_0-unquantized-hereticMoEF1625.8B47.04 GiB0.13 GiB47.70 GiB0.30 GiB7±8.3%
G4-MeroMero-26B-A4B-it-uncensored-hereticMoEBF1625.8B47.04 GiB0.13 GiB47.70 GiB0.30 GiB7±8.3%
gemma-4-26B-A4B-itMoEF1626.5B47.04 GiB0.13 GiB47.70 GiB0.30 GiB7±8.3%
gemma-4-26B-A4B-it-ultra-uncensored-hereticMoEBF1625.8B47.04 GiB0.13 GiB47.70 GiB0.30 GiB7±8.3%
gemma-4-26B-A4B-it-uncensored-hereticMoEBF1625.8B47.04 GiB0.13 GiB47.70 GiB0.30 GiB7±8.3%
gemma-4-26B-A4B-it-abliterixMoEF1625.8B47.04 GiB0.13 GiB47.70 GiB0.30 GiB7±8.3%
gemma-4-26B-A4BMoEBF1626.5B47.04 GiB0.13 GiB47.70 GiB0.30 GiB7±8.3%
gemma-4-26B-A4B-Heretic-StableMoEBF1625.8B47.04 GiB0.13 GiB47.70 GiB0.30 GiB7±8.3%
gemma-4-26B-A4B-it-Uncensored-MAXMoEBF1625.8B47.04 GiB0.13 GiB47.70 GiB0.30 GiB7±8.3%
GLM-4.5-Air-DerestrictedMoEIQ3_XXS110B46.89 GiB0.20 GiB47.67 GiB0.33 GiB29±37%
GLM-4.5-AirMoEIQ3_XXS110B46.89 GiB0.20 GiB47.67 GiB0.33 GiB29±37%
Llama-4-Scout-17B-16E-InstructMoEKV unresolvedIQ3_M109B46.87 GiB0.21 GiB47.65 GiB0.35 GiB29±37%
HunyuanImage-2.1Q5_017.5B47.04 GiB0.00 GiB47.64 GiB0.36 GiB7±8.3%
GLM-4.7-REAP-218B-A32BMoEIQ1_M218B46.56 GiB0.40 GiB47.56 GiB0.44 GiB25±37%
Meta-Llama-3-70B-InstructQ5_K_M70.6B46.53 GiB0.35 GiB47.55 GiB0.45 GiB7±8.3%
Maenad-70BI1-Q5_K_M70.6B46.52 GiB0.35 GiB47.55 GiB0.45 GiB7±8.3%
DeepSeek-R1-Distill-Llama-70B-Uncensored-v2-Unbiased-ReasonerI1-Q5_K_M70.6B46.52 GiB0.35 GiB47.55 GiB0.45 GiB7±8.3%
calme-2.4-llama3-70bQ5_K_M70.6B46.52 GiB0.35 GiB47.55 GiB0.45 GiB7±8.3%
calme-2.2-llama3-70bQ5_K_M70.6B46.52 GiB0.35 GiB47.55 GiB0.45 GiB7±8.3%
Rombos-LLM-70b-Llama-3.3I1-Q5_K_M70.6B46.52 GiB0.35 GiB47.55 GiB0.45 GiB7±8.3%
L3.3-Electra-R1-70bI1-Q5_K_M70.6B46.52 GiB0.35 GiB47.55 GiB0.45 GiB7±8.3%
L3.3-70B-Magnum-v4-SEQ5_K_M70.6B46.52 GiB0.35 GiB47.55 GiB0.45 GiB7±8.3%
Latxa-Llama-3.1-70B-Instruct-v2I1-Q5_K_M70.6B46.52 GiB0.35 GiB47.55 GiB0.45 GiB7±8.3%
Llama-3.3_70_b_uncensored_continuedI1-Q5_K_M70.6B46.52 GiB0.35 GiB47.55 GiB0.45 GiB7±8.3%
Llama-3.3-70B-Instruct-abliteratedI1-Q5_K_M70.6B46.52 GiB0.35 GiB47.55 GiB0.45 GiB7±8.3%
Strawberrylemonade-L3-70B-v1.2Q5_K_M70.6B46.52 GiB0.35 GiB47.55 GiB0.45 GiB7±8.3%
grok-oss-Revenant-70BI1-Q5_K_M70.6B46.52 GiB0.35 GiB47.55 GiB0.45 GiB7±8.3%
Llama-3.1-Nemotron-70B-Instruct-HFI1-Q5_K_M70.6B46.52 GiB0.35 GiB47.55 GiB0.45 GiB7±8.3%
L3.3-70B-Euryale-v2.3I1-Q5_K_M70.6B46.52 GiB0.35 GiB47.55 GiB0.45 GiB7±8.3%
Hermes-4-70B-hereticI1-Q5_K_M70.6B46.52 GiB0.35 GiB47.55 GiB0.45 GiB7±8.3%
Llama-3.3-70B-InstructQ5_K_M70.6B46.52 GiB0.35 GiB47.55 GiB0.45 GiB7±8.3%
Hermes-4-70BQ5_K_M70.6B46.52 GiB0.35 GiB47.55 GiB0.45 GiB7±8.3%
Llama-3.1-70BQ5_K_M70.6B46.52 GiB0.35 GiB47.55 GiB0.45 GiB7±8.3%
Hermes-3-Llama-3.1-70BQ5_K_M70.6B46.52 GiB0.35 GiB47.55 GiB0.45 GiB7±8.3%
Anubis-70B-v1.2Q5_K_M70.6B46.52 GiB0.35 GiB47.55 GiB0.45 GiB7±8.3%
Golem-70B-v1bI1-Q5_K_M70.6B46.52 GiB0.35 GiB47.55 GiB0.45 GiB7±8.3%
DeepSeek-R1-Distill-Llama-70B-abliteratedI1-Q5_K_M70.6B46.52 GiB0.35 GiB47.55 GiB0.45 GiB7±8.3%
DeepSeek-R1-Distill-Llama-70B-hereticI1-Q5_K_M70.6B46.52 GiB0.35 GiB47.55 GiB0.45 GiB7±8.3%
DeepSeek-R1-Distill-Llama-70BQ5_K_M70.6B46.52 GiB0.35 GiB47.55 GiB0.45 GiB7±8.3%
llama-3-firefunction-v2Q5_K_M70.6B46.52 GiB0.35 GiB47.55 GiB0.45 GiB7±8.3%
Legion-V2.1-LLaMa-70BI1-Q5_K_M70.6B46.52 GiB0.35 GiB47.55 GiB0.45 GiB7±8.3%
Tess-R1-Limerick-Llama-3.1-70BQ5_K_M70.6B46.52 GiB0.35 GiB47.55 GiB0.45 GiB7±8.3%
Assistant_Pepe_70BI1-Q5_K_M70.6B46.52 GiB0.35 GiB47.55 GiB0.45 GiB7±8.3%
SEMIKONG-70BQ5_K_M70.6B46.52 GiB0.35 GiB47.55 GiB0.45 GiB7±8.3%
functionary-medium-v3.2KV unresolvedQ5_K_M70.6B46.52 GiB0.35 GiB47.55 GiB0.45 GiB7±8.3%
Llama-3.1-WhiteRabbitNeo-2-70BQ5_K_M70.6B46.52 GiB0.35 GiB47.55 GiB0.45 GiB7±8.3%
Infinity-Instruct-7M-Gen-Llama3_1-70BI1-Q5_K_M70.6B46.52 GiB0.35 GiB47.55 GiB0.45 GiB7±8.3%
Athene-70BQ5_K_M70.6B46.52 GiB0.35 GiB47.55 GiB0.45 GiB7±8.3%
New-Dawn-Llama-3-70B-32K-v1.0I1-Q5_K_M70.6B46.52 GiB0.35 GiB47.55 GiB0.45 GiB7±8.3%
Meta-Llama-3-70B-Instruct-abliterated-v3.5I1-Q5_K_M70.6B46.52 GiB0.35 GiB47.55 GiB0.45 GiB7±8.3%
L3.3-70B-Magnum-DiamondQ5_K_M70.6B46.52 GiB0.35 GiB47.55 GiB0.45 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?
2051 of 2118 indexed open-weight models fit a Apple M2 Max at 4,096 context with q4_0 KV cache, the largest being Hunyuan-A13B-Instruct at 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.