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 16K context with q8_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 16K context

largest quantization that fits, per model · 2089 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
grok-2MoEQ2_K270B93.18 GiB2.13 GiB96.00 GiB0.00 GiB10±37%
Laguna-S-2.1MoEQ6_K_L118B94.83 GiB0.47 GiB95.88 GiB0.12 GiB26±37%
DeepSeek-Coder-V2-Instruct-0724MoEIQ3_S236B94.70 GiB0.56 GiB95.85 GiB0.15 GiB26±37%
DeepSeek-V2.5MoEIQ3_S236B94.70 GiB0.56 GiB95.85 GiB0.15 GiB26±37%
DeepSeek-Coder-V2-InstructMoEIQ3_S236B94.70 GiB0.56 GiB95.85 GiB0.15 GiB26±37%
GLM-4.6-REAP-268B-A32BMoEQ2_K_L269B92.11 GiB3.05 GiB95.75 GiB0.25 GiB19±37%
MiniMax-M2.1MoEI1-IQ3_M229B93.13 GiB2.06 GiB95.72 GiB0.28 GiB25±37%
MiniMax-M2.5MoEI1-IQ3_M229B93.13 GiB2.06 GiB95.72 GiB0.28 GiB25±37%
Mixtral-8x22B-Instruct-v0.1MoEQ5_K_M141B93.11 GiB1.86 GiB95.58 GiB0.42 GiB10±37%
Mixtral-8x22B-v0.1MoEQ5_K_M141B93.11 GiB1.86 GiB95.58 GiB0.42 GiB10±37%
Mixtral-8x22B-v0.1MoEQ5_K_M141B93.10 GiB1.86 GiB95.57 GiB0.43 GiB10±37%
MiMo-V2-FlashMoEKV unresolvedI1-IQ2_M310B93.87 GiB1.00 GiB95.47 GiB0.53 GiB27±37%
MiniMax-M3MoEIQ1_M427B93.82 GiB1.00 GiB95.38 GiB0.62 GiB27±37%
MiniMax-M2.7MoEQ3_K_S229B92.49 GiB2.06 GiB95.09 GiB0.91 GiB25±37%
gemma-4-26B-A4B-it-Uncensored-MAXMoEF3225.8B94.02 GiB0.49 GiB95.04 GiB0.96 GiB5±8.3%
Step-3.7-FlashUD-IQ4_NL201B90.63 GiB3.74 GiB94.95 GiB1.05 GiB5±8.3%
MiniMax-M2MoEQ3_K_S229B92.31 GiB2.06 GiB94.90 GiB1.10 GiB25±37%
Llama-4-Maverick-17B-128E-InstructMoEKV unresolvedTQ1_0402B92.59 GiB1.59 GiB94.76 GiB1.24 GiB31±37%
MiniMax-M2.1-REAP-139B-A10BMoEI1-Q5_K_M139B91.98 GiB2.06 GiB94.57 GiB1.43 GiB22±37%
m51Lab-MiniMax-M2.7-REAP-139B-A10BMoEI1-Q5_K_M139B91.98 GiB2.06 GiB94.57 GiB1.43 GiB22±37%
MiniMax-M2.7-BF16-ultra-uncensored-hereticMoEI1-IQ3_S229B91.94 GiB2.06 GiB94.53 GiB1.47 GiB25±37%
GLM-4.5-Air-DerestrictedMoEQ6_K110B92.37 GiB1.53 GiB94.48 GiB1.52 GiB21±37%
GLM-4.5-AirMoEQ6_K110B92.37 GiB1.53 GiB94.48 GiB1.52 GiB21±37%
Qwen3.5-122B-A10B-hereticMoEI1-Q6_K123B93.42 GiB0.20 GiB94.20 GiB1.80 GiB29±37%
GLM-4.7MoEUD-IQ1_S358B90.50 GiB3.05 GiB94.15 GiB1.85 GiB21±37%
GLM-4.7-REAP-218B-A32BMoEQ3_K_S218B90.39 GiB3.05 GiB94.03 GiB1.97 GiB17±37%
GLM-4.5MoEUD-IQ1_S358B90.38 GiB3.05 GiB94.03 GiB1.97 GiB21±37%
GLM-4.6MoEUD-IQ1_S357B90.28 GiB3.05 GiB93.93 GiB2.07 GiB21±37%
DeepSeek-V4-FlashMoEQ2_K291B92.86 GiB0.03 GiB93.49 GiB2.51 GiB32±37%
Mistral-Small-4-119B-2603MoEUD-Q6_K119B92.60 GiB0.19 GiB93.37 GiB2.63 GiB29±37%
Hermes-4-405BUD-IQ1_M406B88.23 GiB4.18 GiB93.24 GiB2.76 GiB6±8.3%
dots.llm1.instMoEQ4_1143B84.24 GiB8.23 GiB93.05 GiB2.95 GiB16±37%
Qwen3-235B-A22B-Instruct-2507MoEIQ3_XS235B90.25 GiB1.56 GiB92.39 GiB3.61 GiB21±37%
Qwen3-235B-A22B-Thinking-2507MoEIQ3_XS235B90.25 GiB1.56 GiB92.39 GiB3.61 GiB21±37%
Qwen3.5-397B-A17BMoEIQ1_M403B91.53 GiB0.25 GiB92.38 GiB3.62 GiB33±37%
Kimi-Linear-48B-A3B-InstructMoEBF1649.1B91.54 GiB0.25 GiB92.35 GiB3.65 GiB6±8.3%
command-a-plus-05-2026-bf16MoEQ3_K_S219B91.28 GiB0.49 GiB92.33 GiB3.67 GiB23±37%
Hermes-3-Llama-3.1-405BIQ1_M406B87.08 GiB4.18 GiB92.09 GiB3.91 GiB6±8.3%
Solar-Open2-250BMoEIQ3_XXS250B89.86 GiB1.59 GiB92.03 GiB3.97 GiB28±37%
GLM-4.6VMoEQ6_K_L108B89.58 GiB1.53 GiB91.69 GiB4.31 GiB22±37%
NVIDIA-Nemotron-3-Super-120B-A12B-BF16MoEQ5_K_L124B90.28 GiB0.73 GiB91.55 GiB4.45 GiB26±37%
MiMo-V2.5MoEKV unresolvedUD-IQ2_M311B89.93 GiB1.00 GiB91.52 GiB4.48 GiB28±37%
Qwen3-235B-A22B-abliteratedMoEI1-IQ3_XS235B89.36 GiB1.56 GiB91.51 GiB4.49 GiB22±37%
GLM-4.5VMoEI1-Q6_K108B89.28 GiB1.53 GiB91.38 GiB4.62 GiB22±37%
Nex-N2-ProMoEIQ1_M397B90.02 GiB0.25 GiB90.87 GiB5.13 GiB33±37%
Qwen3-VL-235B-A22B-ThinkingMoEUD-IQ3_XXS236B88.22 GiB1.56 GiB90.37 GiB5.63 GiB22±37%
Qwen3-VL-235B-A22B-InstructMoEUD-IQ3_XXS236B88.15 GiB1.56 GiB90.29 GiB5.71 GiB22±37%
Trinity-Large-PreviewMoEIQ2_XXS399B88.58 GiB0.92 GiB90.07 GiB5.93 GiB32±37%
Trinity-Large-TrueBaseMoEIQ2_XXS399B88.58 GiB0.92 GiB90.07 GiB5.93 GiB32±37%
Dolphin3.0-R1-Mistral-24BF3223.6B87.82 GiB1.33 GiB89.82 GiB6.18 GiB6±8.3%
Dolphin3.0-Mistral-24BF3223.6B87.82 GiB1.33 GiB89.82 GiB6.18 GiB6±8.3%
Mistral-Small-24B-Instruct-2501-abliteratedF3223.6B87.82 GiB1.33 GiB89.82 GiB6.18 GiB6±8.3%
Mistral-Small-24B-Instruct-2501F3223.6B87.82 GiB1.33 GiB89.82 GiB6.18 GiB6±8.3%
Hy3MoEIQ2_S299B86.43 GiB2.66 GiB89.67 GiB6.33 GiB23±37%
Mistral-Medium-3.5-128BQ5_K_L128B85.78 GiB2.92 GiB89.41 GiB6.59 GiB6±8.3%
gpt-oss-20b-hereticMoEQ5_120.9B88.57 GiB0.21 GiB89.31 GiB6.69 GiB17±37%
Huihui-gpt-oss-20b-BF16-abliteratedMoEQ5_120.9B88.57 GiB0.21 GiB89.31 GiB6.69 GiB17±37%
ERNIE-4.5-300B-A47B-PTUD-IQ1_M300B86.78 GiB1.79 GiB89.25 GiB6.75 GiB6±8.3%
Qwen3.6-27B-Fable-Fusion-711-Uncensored-Heretic-NM-DAU-MTPQ6_K27.8B88.00 GiB0.53 GiB89.15 GiB6.85 GiB6±8.3%
xLAM-8x7b-rMoEBF1646.7B86.99 GiB1.06 GiB88.64 GiB7.36 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 16,384 context with q8_0 KV cache, the largest being grok-2 at Q2_K. 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.