Apple · apple

Apple M5 Max

Apple M5 Max has 128 GB of unified memory at 614 GB/s — about 89.28 GiB usable after driver and compositor overhead. 2089 of 2118 indexed models fit at 32K context with q4_0 KV. Note only 96 GB of its 128 GB is allocatable to the GPU.

Spec sheet· bandwidth, theoreticalFrom the file· fit from summed bytesPredicted· speed
Memory
128 GB
LPDDR5X-9600
Bandwidth
614 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 1794vision language 191audio tts 21image 2audio asr 39video 16embedding 26

What fits at 32K context

largest quantization that fits, per model · 2089 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
GLM-4.6-REAP-268B-A32BMoEQ2_K_L269B92.11 GiB3.23 GiB95.93 GiB0.07 GiB18±37%
DeepSeek-Coder-V2-Instruct-0724MoEIQ3_S236B94.70 GiB0.59 GiB95.88 GiB0.12 GiB26±37%
DeepSeek-V2.5MoEIQ3_S236B94.70 GiB0.59 GiB95.88 GiB0.12 GiB26±37%
DeepSeek-Coder-V2-InstructMoEIQ3_S236B94.70 GiB0.59 GiB95.88 GiB0.12 GiB26±37%
Laguna-S-2.1MoEQ6_K_L118B94.83 GiB0.46 GiB95.87 GiB0.13 GiB26±37%
MiniMax-M2.1MoEI1-IQ3_M229B93.13 GiB2.18 GiB95.84 GiB0.16 GiB24±37%
MiniMax-M2.5MoEI1-IQ3_M229B93.13 GiB2.18 GiB95.84 GiB0.16 GiB24±37%
Mixtral-8x22B-Instruct-v0.1MoEQ5_K_M141B93.11 GiB1.97 GiB95.69 GiB0.31 GiB10±37%
Mixtral-8x22B-v0.1MoEQ5_K_M141B93.11 GiB1.97 GiB95.69 GiB0.31 GiB10±37%
Mixtral-8x22B-v0.1MoEQ5_K_M141B93.10 GiB1.97 GiB95.68 GiB0.32 GiB10±37%
MiMo-V2-FlashMoEKV unresolvedI1-IQ2_M310B93.87 GiB1.05 GiB95.53 GiB0.47 GiB27±37%
grok-2MoEUD-IQ2_XXS270B92.52 GiB2.25 GiB95.47 GiB0.53 GiB10±37%
MiniMax-M3MoEIQ1_M427B93.82 GiB1.05 GiB95.44 GiB0.56 GiB27±37%
MiniMax-M2.7MoEQ3_K_S229B92.49 GiB2.18 GiB95.21 GiB0.79 GiB25±37%
MiniMax-M2MoEQ3_K_S229B92.31 GiB2.18 GiB95.02 GiB0.98 GiB25±37%
gemma-4-26B-A4B-it-Uncensored-MAXMoEF3225.8B94.02 GiB0.43 GiB94.99 GiB1.01 GiB5±8.3%
Step-3.7-FlashUD-IQ4_NL201B90.63 GiB3.67 GiB94.88 GiB1.12 GiB5±8.3%
Llama-4-Maverick-17B-128E-InstructMoEKV unresolvedTQ1_0402B92.59 GiB1.69 GiB94.86 GiB1.14 GiB31±37%
MiniMax-M2.1-REAP-139B-A10BMoEI1-Q5_K_M139B91.98 GiB2.18 GiB94.69 GiB1.31 GiB22±37%
m51Lab-MiniMax-M2.7-REAP-139B-A10BMoEI1-Q5_K_M139B91.98 GiB2.18 GiB94.69 GiB1.31 GiB22±37%
MiniMax-M2.7-BF16-ultra-uncensored-hereticMoEI1-IQ3_S229B91.94 GiB2.18 GiB94.65 GiB1.35 GiB25±37%
GLM-4.5-Air-DerestrictedMoEQ6_K110B92.37 GiB1.62 GiB94.57 GiB1.43 GiB21±37%
GLM-4.5-AirMoEQ6_K110B92.37 GiB1.62 GiB94.57 GiB1.43 GiB21±37%
GLM-4.7MoEUD-IQ1_S358B90.50 GiB3.23 GiB94.33 GiB1.67 GiB20±37%
GLM-4.7-REAP-218B-A32BMoEQ3_K_S218B90.39 GiB3.23 GiB94.21 GiB1.79 GiB17±37%
Qwen3.5-122B-A10B-hereticMoEI1-Q6_K123B93.42 GiB0.21 GiB94.21 GiB1.79 GiB29±37%
GLM-4.5MoEUD-IQ1_S358B90.38 GiB3.23 GiB94.21 GiB1.79 GiB20±37%
GLM-4.6MoEUD-IQ1_S357B90.28 GiB3.23 GiB94.11 GiB1.89 GiB21±37%
dots.llm1.instMoEQ4_1143B84.24 GiB8.72 GiB93.53 GiB2.47 GiB16±37%
Hermes-4-405BUD-IQ1_M406B88.23 GiB4.43 GiB93.49 GiB2.51 GiB6±8.3%
DeepSeek-V4-FlashMoEQ2_K291B92.86 GiB0.02 GiB93.47 GiB2.53 GiB32±37%
Mistral-Small-4-119B-2603MoEUD-Q6_K119B92.60 GiB0.20 GiB93.38 GiB2.62 GiB29±37%
Qwen3-235B-A22B-Instruct-2507MoEIQ3_XS235B90.25 GiB1.65 GiB92.48 GiB3.52 GiB21±37%
Qwen3-235B-A22B-Thinking-2507MoEIQ3_XS235B90.25 GiB1.65 GiB92.48 GiB3.52 GiB21±37%
Qwen3.5-397B-A17BMoEIQ1_M403B91.53 GiB0.26 GiB92.40 GiB3.60 GiB33±37%
Kimi-Linear-48B-A3B-InstructMoEBF1649.1B91.54 GiB0.27 GiB92.36 GiB3.64 GiB6±8.3%
Hermes-3-Llama-3.1-405BIQ1_M406B87.08 GiB4.43 GiB92.34 GiB3.66 GiB6±8.3%
command-a-plus-05-2026-bf16MoEQ3_K_S219B91.28 GiB0.40 GiB92.24 GiB3.76 GiB23±37%
Solar-Open2-250BMoEIQ3_XXS250B89.86 GiB1.69 GiB92.12 GiB3.88 GiB27±37%
GLM-4.6VMoEQ6_K_L108B89.58 GiB1.62 GiB91.78 GiB4.22 GiB21±37%
Qwen3-235B-A22B-abliteratedMoEI1-IQ3_XS235B89.36 GiB1.65 GiB91.60 GiB4.40 GiB21±37%
NVIDIA-Nemotron-3-Super-120B-A12B-BF16MoEQ5_K_L124B90.28 GiB0.77 GiB91.60 GiB4.40 GiB26±37%
MiMo-V2.5MoEKV unresolvedUD-IQ2_M311B89.93 GiB1.05 GiB91.58 GiB4.42 GiB28±37%
GLM-4.5VMoEI1-Q6_K108B89.28 GiB1.62 GiB91.47 GiB4.53 GiB21±37%
Nex-N2-ProMoEIQ1_M397B90.02 GiB0.26 GiB90.88 GiB5.12 GiB33±37%
Qwen3-VL-235B-A22B-ThinkingMoEUD-IQ3_XXS236B88.22 GiB1.65 GiB90.46 GiB5.54 GiB22±37%
Qwen3-VL-235B-A22B-InstructMoEUD-IQ3_XXS236B88.15 GiB1.65 GiB90.38 GiB5.62 GiB22±37%
Trinity-Large-PreviewMoEIQ2_XXS399B88.58 GiB0.75 GiB89.90 GiB6.10 GiB33±37%
Trinity-Large-TrueBaseMoEIQ2_XXS399B88.58 GiB0.75 GiB89.90 GiB6.10 GiB33±37%
Dolphin3.0-R1-Mistral-24BF3223.6B87.82 GiB1.41 GiB89.90 GiB6.10 GiB6±8.3%
Dolphin3.0-Mistral-24BF3223.6B87.82 GiB1.41 GiB89.90 GiB6.10 GiB6±8.3%
Mistral-Small-24B-Instruct-2501-abliteratedF3223.6B87.82 GiB1.41 GiB89.90 GiB6.10 GiB6±8.3%
Mistral-Small-24B-Instruct-2501F3223.6B87.82 GiB1.41 GiB89.90 GiB6.10 GiB6±8.3%
Hy3MoEIQ2_S299B86.43 GiB2.81 GiB89.83 GiB6.17 GiB23±37%
Mistral-Medium-3.5-128BQ5_K_L128B85.78 GiB3.09 GiB89.58 GiB6.42 GiB6±8.3%
ERNIE-4.5-300B-A47B-PTUD-IQ1_M300B86.78 GiB1.90 GiB89.35 GiB6.65 GiB6±8.3%
gpt-oss-20b-hereticMoEQ5_120.9B88.57 GiB0.22 GiB89.32 GiB6.68 GiB17±37%
Huihui-gpt-oss-20b-BF16-abliteratedMoEQ5_120.9B88.57 GiB0.22 GiB89.32 GiB6.68 GiB17±37%
Qwen3.6-27B-Fable-Fusion-711-Uncensored-Heretic-NM-DAU-MTPQ6_K27.8B88.00 GiB0.56 GiB89.18 GiB6.82 GiB6±8.3%
xLAM-8x7b-rMoEBF1646.7B86.99 GiB1.13 GiB88.71 GiB7.29 GiB11±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 M5 Max run?
2089 of 2118 indexed open-weight models fit a Apple M5 Max at 32,768 context with q4_0 KV cache, the largest being GLM-4.6-REAP-268B-A32B at Q2_K_L. That covers text, vision-language, image, video and speech models.
How much usable memory does a Apple M5 Max actually have?
Its nameplate is 128 GB, but about 89.28 GiB is available to a model once driver and compositor overhead is accounted for, and only 96 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 614 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.