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. 1940 of 2118 indexed models fit at 64K 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 1661vision language 175video 16image 2audio asr 39audio tts 21embedding 26

What fits at 64K context

largest quantization that fits, per model · 1940 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
MN-GRAND-23.5B-Gutenberg-UNCENSORED-V2-GLM4.7-ThinkingQ5_K_M23.4B15.64 GiB10.76 GiB27.00 GiB0.00 GiB14±8.3%
Hunyuan-A13B-InstructMoEIQ2_S80.4B22.19 GiB4.25 GiB26.99 GiB0.01 GiB14±8.3%
glm-4-9b-chat-1mQ4_09.5B5.14 GiB21.25 GiB26.99 GiB0.01 GiB14±8.3%
Phi-3.5-MoE-instructMoEKV unresolvedQ4_K_S41.9B22.18 GiB4.25 GiB26.98 GiB0.02 GiB24±37%
Seed-OSS-36B-Instruct-biprojected-norm-preserving-abliteratedI1-Q3_K_L36.2B17.83 GiB8.50 GiB26.98 GiB0.02 GiB14±8.3%
Seed-OSS-36B-InstructQ3_K_L36.2B17.83 GiB8.50 GiB26.98 GiB0.02 GiB14±8.3%
Hermes-4.3-36B-hereticI1-Q3_K_L36.2B17.83 GiB8.50 GiB26.98 GiB0.02 GiB14±8.3%
Hermes-4.3-36BQ3_K_L36.2B17.83 GiB8.50 GiB26.98 GiB0.02 GiB14±8.3%
Qwen3-Coder-NextMoEUD-IQ2_M79.7B23.25 GiB3.19 GiB26.98 GiB0.02 GiB35±37%
glm-4-9b-chatIQ4_NL9.4B5.13 GiB21.25 GiB26.97 GiB0.03 GiB14±8.3%
SOLAR-10.7B-Instruct-v1.0F1610.7B19.99 GiB6.38 GiB26.95 GiB0.05 GiB14±8.3%
glm-4-9b-chat-abliteratedQ4_09.4B5.10 GiB21.25 GiB26.94 GiB0.06 GiB14±8.3%
codegeex4-all-9bIQ4_NL9.4B5.08 GiB21.25 GiB26.93 GiB0.07 GiB14±8.3%
c4ai-command-r-08-2024Q5_K_S32.3B20.95 GiB5.31 GiB26.93 GiB0.07 GiB14±8.3%
Qwen3-48B-A4B-Savant-Commander-Distill-12X-Closed-Open-Heretic-UncensoredMoEI1-Q5_K_S33.6B21.58 GiB4.78 GiB26.92 GiB0.08 GiB21±37%
deepseek-llm-67b-chatI1-IQ1_S67.4B13.65 GiB12.62 GiB26.92 GiB0.08 GiB14±8.3%
deepseek-llm-67b-baseI1-IQ1_S67.4B13.65 GiB12.62 GiB26.92 GiB0.08 GiB14±8.3%
openbuddy-deepseek-67b-v15.3-4kI1-IQ1_S67.4B13.65 GiB12.62 GiB26.92 GiB0.08 GiB14±8.3%
Gemma-4-Gembrain-X-Core-31BI1-Q5_K_M31.3B20.35 GiB5.94 GiB26.91 GiB0.09 GiB14±8.3%
Gemma-4-Gembrain-X-31BI1-Q5_K_M31.3B20.35 GiB5.94 GiB26.91 GiB0.09 GiB14±8.3%
Gemma-4-31B-Isometry-Fabled-PersonaI1-Q5_K_M31.3B20.35 GiB5.94 GiB26.91 GiB0.09 GiB14±8.3%
Versipellis-31BI1-Q5_K_M31.3B20.35 GiB5.94 GiB26.91 GiB0.09 GiB14±8.3%
Gemma4-Gutenberg-31BI1-Q5_K_M31.3B20.35 GiB5.94 GiB26.91 GiB0.09 GiB14±8.3%
G4-MeroMero-31B-uncensored-hereticI1-Q5_K_M31.3B20.35 GiB5.94 GiB26.91 GiB0.09 GiB14±8.3%
Gemma-4-Novelist-31BI1-Q5_K_M31.3B20.35 GiB5.94 GiB26.91 GiB0.09 GiB14±8.3%
Wanabi-Gemma4-31BI1-Q5_K_M31.3B20.35 GiB5.94 GiB26.91 GiB0.09 GiB14±8.3%
G4-Alice-v1.2-31BI1-Q5_K_M31.3B20.35 GiB5.94 GiB26.91 GiB0.09 GiB14±8.3%
Agares-31B-v1I1-Q5_K_M30.7B20.35 GiB5.94 GiB26.91 GiB0.09 GiB14±8.3%
Gemma4-Gutenberg-31B-HereticI1-Q5_K_M31.3B20.35 GiB5.94 GiB26.91 GiB0.09 GiB14±8.3%
gemma-4-Ortenzya-The-Creative-Wordsmith-31B-it-uncensored-hereticI1-Q5_K_M31.3B20.35 GiB5.94 GiB26.91 GiB0.09 GiB14±8.3%
Gemma-4-Gemsicle-31BI1-Q5_K_M31.3B20.35 GiB5.94 GiB26.91 GiB0.09 GiB14±8.3%
Gemma-4-Gembrain-31B-it-uncensored-hereticI1-Q5_K_M31.3B20.35 GiB5.94 GiB26.91 GiB0.09 GiB14±8.3%
Melinoe-Gemma4-31B-VL-hereticI1-Q5_K_M31.3B20.35 GiB5.94 GiB26.91 GiB0.09 GiB14±8.3%
G4-MeroMero-31BI1-Q5_K_M31.3B20.35 GiB5.94 GiB26.91 GiB0.09 GiB14±8.3%
Glistening-Gem-31B-v1.0I1-Q5_K_M31.3B20.35 GiB5.94 GiB26.91 GiB0.09 GiB14±8.3%
Melinoe-Gemma4-31B-VLI1-Q5_K_M31.3B20.35 GiB5.94 GiB26.91 GiB0.09 GiB14±8.3%
Gemma-4-31B-Storymaxxed3I1-Q5_K_M31.3B20.35 GiB5.94 GiB26.91 GiB0.09 GiB14±8.3%
Gemma-4-AssGuard-31BI1-Q5_K_M31.3B20.35 GiB5.94 GiB26.91 GiB0.09 GiB14±8.3%
copywriter-gemma4-31bI1-Q5_K_M32.7B20.35 GiB5.94 GiB26.91 GiB0.09 GiB14±8.3%
gemma-4-31B-heretic-finetuneI1-Q5_K_M30.7B20.35 GiB5.94 GiB26.91 GiB0.09 GiB14±8.3%
Gemma-4-Garnet-V2-31B-it-ultra-uncensored-hereticI1-Q5_K_M31.3B20.35 GiB5.94 GiB26.91 GiB0.09 GiB14±8.3%
gemma-4-31B-it-Claude-Opus-Distill-v2Q5_K_M32.7B20.35 GiB5.94 GiB26.91 GiB0.09 GiB14±8.3%
gemma-4-31B-it-abliterated-v3I1-Q5_K_M31.3B20.35 GiB5.94 GiB26.91 GiB0.09 GiB14±8.3%
Gemma-4-Harmonia-31B-uncensored-hereticQ5_K_M31.3B20.35 GiB5.94 GiB26.91 GiB0.09 GiB14±8.3%
gemma-4-31B-it-noloopI1-Q5_K_M31.3B20.35 GiB5.94 GiB26.91 GiB0.09 GiB14±8.3%
Webs-Sejong-31B-v7I1-Q5_K_M31.3B20.35 GiB5.94 GiB26.91 GiB0.09 GiB14±8.3%
Lilith-31B-v1.0I1-Q5_K_M31.3B20.35 GiB5.94 GiB26.91 GiB0.09 GiB14±8.3%
JGOS-31B-ThinkI1-Q5_K_M31.3B20.35 GiB5.94 GiB26.91 GiB0.09 GiB14±8.3%
gemma-4-31B-MergemaxxedI1-Q5_K_M31.3B20.35 GiB5.94 GiB26.91 GiB0.09 GiB14±8.3%
K1-v6-zeroI1-Q5_K_M32.7B20.35 GiB5.94 GiB26.91 GiB0.09 GiB14±8.3%
gemma-4-31B-it-uncensored-hereticQ5_K_M31.3B20.35 GiB5.94 GiB26.91 GiB0.09 GiB14±8.3%
Gemma-4-Queen-31B-it-uncensored-hereticI1-Q5_K_M31.3B20.35 GiB5.94 GiB26.91 GiB0.09 GiB14±8.3%
Gemma-4-Sphinsikus-Chronist-31BI1-Q5_K_M31.3B20.35 GiB5.94 GiB26.91 GiB0.09 GiB14±8.3%
gemma-4-31B-it-hereticI1-Q5_K_M31.3B20.35 GiB5.94 GiB26.91 GiB0.09 GiB14±8.3%
Gemma4-31B-Finetuned-V2I1-Q5_K_M32.7B20.35 GiB5.94 GiB26.91 GiB0.09 GiB14±8.3%
Gemma-4-31B-storymaxxedI1-Q5_K_M31.3B20.35 GiB5.94 GiB26.91 GiB0.09 GiB14±8.3%
Gemma-4-31B-Fable-5-Agent-DistillQ5_K_M32.7B20.35 GiB5.94 GiB26.91 GiB0.09 GiB14±8.3%
gemma-4-31B-it-uncensoredQ5_K_M31.3B20.35 GiB5.94 GiB26.91 GiB0.09 GiB14±8.3%
Gemma-4-31B-storymaxxed2I1-Q5_K_M31.3B20.35 GiB5.94 GiB26.91 GiB0.09 GiB14±8.3%
gemma-4-31B-it-abliteratedQ5_K_M31.3B20.35 GiB5.94 GiB26.91 GiB0.09 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?
1940 of 2118 indexed open-weight models fit a Apple M5 Max at 65,536 context with q8_0 KV cache, the largest being MN-GRAND-23.5B-Gutenberg-UNCENSORED-V2-GLM4.7-Thinking at Q5_K_M. 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.