Apple · apple

Apple M5 Max

Apple M5 Max has 36 GB of unified memory at 461 GB/s — about 25.11 GiB usable after driver and compositor overhead. 1973 of 2118 indexed models fit at 32K context with q8_0 KV. Note only 27 GB of its 36 GB is allocatable to the GPU.

Spec sheet· bandwidth, theoreticalFrom the file· fit from summed bytesPredicted· speed
Memory
36 GB
LPDDR5X-9600
Bandwidth
461 GB/s
384-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 1693vision language 176audio tts 21video 16image 2audio asr 39embedding 26

What fits at 32K context

largest quantization that fits, per model · 1973 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
IQuest-Coder-V1-40B-InstructI1-Q4_039.8B21.03 GiB5.31 GiB26.99 GiB0.01 GiB14±8.3%
spoomplesmaxx-v2.1-30BI1-Q6_K28.9B22.06 GiB4.25 GiB26.97 GiB0.03 GiB14±8.3%
Huihui-granite-4.1-30b-abliteratedI1-Q6_K28.9B22.06 GiB4.25 GiB26.97 GiB0.03 GiB14±8.3%
granite-4.1-30b-hereticI1-Q6_K28.9B22.06 GiB4.25 GiB26.97 GiB0.03 GiB14±8.3%
granite-4.1-30bQ6_K28.9B22.06 GiB4.25 GiB26.97 GiB0.03 GiB14±8.3%
GLM-Z1-32B-0414Q6_K_L32.6B25.31 GiB1.01 GiB26.96 GiB0.04 GiB14±8.3%
GLM-4-32B-0414Q6_K_L32.6B25.31 GiB1.01 GiB26.96 GiB0.04 GiB14±8.3%
OpenBuddy-R1-0528-Distill-Qwen3-32B-Preview0-QATQ5_K_L32.8B22.07 GiB4.25 GiB26.96 GiB0.04 GiB14±8.3%
KAT-DevQ5_K_L32.8B22.07 GiB4.25 GiB26.96 GiB0.04 GiB14±8.3%
Qwen3-VL-32B-InstructQ5_K_L33.4B22.07 GiB4.25 GiB26.96 GiB0.04 GiB14±8.3%
DeepSWE-PreviewQ5_K_L32.8B22.07 GiB4.25 GiB26.96 GiB0.04 GiB14±8.3%
Huihui-Qwen3-Coder-Next-abliteratedMoEQ2_K79.7B26.00 GiB0.40 GiB26.94 GiB0.06 GiB59±37%
Apertus-70B-Instruct-2509IQ2_S70.6B20.89 GiB5.31 GiB26.94 GiB0.06 GiB14±8.3%
Assistant_Pepe_70BIQ2_XXS70.6B20.94 GiB5.31 GiB26.93 GiB0.07 GiB14±8.3%
MythoMax-L2-Kimiko-v2-13bQ8_013.0B13.02 GiB13.28 GiB26.90 GiB0.10 GiB14±8.3%
Salience-1.5-FlashMoEQ6_K_L31.1B24.76 GiB1.59 GiB26.90 GiB0.10 GiB38±37%
Qwen3.6-27B-A3B-CoderMoEQ8_026.7B25.99 GiB0.33 GiB26.88 GiB0.12 GiB49±37%
Hypernova-60B-2605MoEI1-IQ2_XXS58.7B25.80 GiB0.54 GiB26.88 GiB0.12 GiB48±37%
EXAONE-4.0-32BQ6_K_L32.0B24.69 GiB1.51 GiB26.85 GiB0.15 GiB14±8.3%
magnum-v2-32bQ5_K_L32.5B21.95 GiB4.25 GiB26.84 GiB0.16 GiB14±8.3%
deepseek-llm-67b-chatI1-IQ2_S67.4B19.88 GiB6.31 GiB26.84 GiB0.16 GiB14±8.3%
deepseek-llm-67b-baseI1-IQ2_S67.4B19.88 GiB6.31 GiB26.84 GiB0.16 GiB14±8.3%
openbuddy-deepseek-67b-v15.3-4kI1-IQ2_S67.4B19.88 GiB6.31 GiB26.84 GiB0.16 GiB14±8.3%
Ace-Step1.5Q5_K_M160M25.46 GiB0.82 GiB26.82 GiB0.18 GiB14±8.3%
Olmo-3.1-32B-InstructQ6_K32.2B24.63 GiB1.51 GiB26.79 GiB0.21 GiB14±8.3%
Olmo-3.1-32B-ThinkQ6_K32.2B24.63 GiB1.51 GiB26.79 GiB0.21 GiB14±8.3%
Olmo-3-32B-ThinkQ6_K32.2B24.63 GiB1.51 GiB26.79 GiB0.21 GiB14±8.3%
Qwen3-Coder-REAP-25B-A3BMoEQ8_024.9B24.64 GiB1.59 GiB26.78 GiB0.22 GiB36±37%
codellama-13b-oasst-sft-v10Q8_013.0B12.88 GiB13.28 GiB26.76 GiB0.24 GiB14±8.3%
chronos-hermes-13b-v2Q8_013.0B12.88 GiB13.28 GiB26.76 GiB0.24 GiB14±8.3%
WhiteRabbitNeo-13B-v1Q8_013.0B12.88 GiB13.28 GiB26.76 GiB0.24 GiB14±8.3%
CodeLlama-13b-Instruct-hfQ8_013.0B12.88 GiB13.28 GiB26.76 GiB0.24 GiB14±8.3%
Orca-2-13b-Alpaca-UncensoredQ8_013.0B12.88 GiB13.28 GiB26.76 GiB0.24 GiB14±8.3%
WizardCoder-Python-13B-V1.0Q8_013.0B12.88 GiB13.28 GiB26.76 GiB0.24 GiB14±8.3%
WizardLM-13B-UncensoredQ8_013.0B12.88 GiB13.28 GiB26.76 GiB0.24 GiB14±8.3%
Llama-2-13b-chat-hfQ8_013.0B12.88 GiB13.28 GiB26.76 GiB0.24 GiB14±8.3%
Wizard-Vicuna-13B-UncensoredQ8_013.0B12.88 GiB13.28 GiB26.76 GiB0.24 GiB14±8.3%
WizardLM-13b-V1.0-UncensoredQ8_013.0B12.88 GiB13.28 GiB26.76 GiB0.24 GiB14±8.3%
Guanaco-13B-UncensoredQ8_013.0B12.88 GiB13.28 GiB26.76 GiB0.24 GiB14±8.3%
MythoMax-L2-13bQ8_013.0B12.88 GiB13.28 GiB26.76 GiB0.24 GiB14±8.3%
WizardLM-1.0-Uncensored-Llama2-13bQ8_013.0B12.88 GiB13.28 GiB26.76 GiB0.24 GiB14±8.3%
speechless-llama2-hermes-orca-platypus-wizardlm-13bQ8_013.0B12.88 GiB13.28 GiB26.76 GiB0.24 GiB14±8.3%
mythalion-13bQ8_013.0B12.88 GiB13.28 GiB26.76 GiB0.24 GiB14±8.3%
ALIA-40b-fc-2606I1-Q4_K_M40.4B22.90 GiB3.19 GiB26.75 GiB0.25 GiB14±8.3%
ALIA-40b-instruct-2606I1-Q4_K_M40.4B22.90 GiB3.19 GiB26.75 GiB0.25 GiB14±8.3%
Trinity-MiniMoEQ8_026.1B25.88 GiB0.33 GiB26.75 GiB0.25 GiB47±37%
Noromaid-v0.4-Mixtral-Instruct-8x7b-ZlossMoEIQ4_XS46.7B24.02 GiB2.13 GiB26.73 GiB0.27 GiB21±37%
Noromaid-20b-v0.1.1I1-IQ2_XS20.0B5.53 GiB20.59 GiB26.71 GiB0.29 GiB14±8.3%
Qwen3-42B-A3B-2507-Thinking-Abliterated-uncensored-TOTAL-RECALL-v2-Medium-MASTER-CODERMoEI1-Q4_K_M42.4B23.94 GiB2.22 GiB26.71 GiB0.29 GiB35±37%
Maenad-70BI1-IQ2_S70.6B20.71 GiB5.31 GiB26.70 GiB0.30 GiB14±8.3%
DeepSeek-R1-Distill-Llama-70B-Uncensored-v2-Unbiased-ReasonerI1-IQ2_S70.6B20.71 GiB5.31 GiB26.70 GiB0.30 GiB14±8.3%
Rombos-LLM-70b-Llama-3.3I1-IQ2_S70.6B20.71 GiB5.31 GiB26.70 GiB0.30 GiB14±8.3%
L3.3-Electra-R1-70bI1-IQ2_S70.6B20.71 GiB5.31 GiB26.70 GiB0.30 GiB14±8.3%
L3.3-70B-Magnum-v4-SEIQ2_S70.6B20.71 GiB5.31 GiB26.70 GiB0.30 GiB14±8.3%
Latxa-Llama-3.1-70B-Instruct-v2I1-IQ2_S70.6B20.71 GiB5.31 GiB26.70 GiB0.30 GiB14±8.3%
Llama-3.3_70_b_uncensored_continuedI1-IQ2_S70.6B20.71 GiB5.31 GiB26.70 GiB0.30 GiB14±8.3%
Llama-3.3-70B-Instruct-abliteratedI1-IQ2_S70.6B20.71 GiB5.31 GiB26.70 GiB0.30 GiB14±8.3%
grok-oss-Revenant-70BI1-IQ2_S70.6B20.71 GiB5.31 GiB26.70 GiB0.30 GiB14±8.3%
Llama-3.1-Nemotron-70B-Instruct-HFI1-IQ2_S70.6B20.71 GiB5.31 GiB26.70 GiB0.30 GiB14±8.3%
L3.3-70B-Euryale-v2.3I1-IQ2_S70.6B20.71 GiB5.31 GiB26.70 GiB0.30 GiB14±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 M5 Max run?
1973 of 2118 indexed open-weight models fit a Apple M5 Max at 32,768 context with q8_0 KV cache, the largest being IQuest-Coder-V1-40B-Instruct at I1-Q4_0. That covers text, vision-language, image, video and speech models.
How much usable memory does a Apple M5 Max actually have?
Its nameplate is 36 GB, but about 25.11 GiB is available to a model once driver and compositor overhead is accounted for, and only 27 GB of the pool can be allocated to the GPU at all.
Is a Apple M5 Max fast for local AI?
Its memory bandwidth is 461 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.