Apple · apple

Apple M2 Max

Apple M2 Max has 96 GB of unified memory at 410 GB/s — about 66.96 GiB usable after driver and compositor overhead. 2071 of 2118 indexed models fit at 64K context with q4_0 KV. Note only 72 GB of its 96 GB is allocatable to the GPU.

Spec sheet· bandwidth, theoreticalFrom the file· fit from summed bytesPredicted· speed
Memory
96 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 1781vision language 186image 2audio asr 39audio tts 21video 16embedding 26

What fits at 64K context

largest quantization that fits, per model · 2071 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
MiniMax-M2.1MoEIQ2_M229B67.05 GiB4.36 GiB71.95 GiB0.05 GiB17±37%
MiniMax-M2MoEIQ2_M229B67.05 GiB4.36 GiB71.95 GiB0.05 GiB17±37%
dots.llm1.instMoEIQ2_M143B53.68 GiB17.44 GiB71.70 GiB0.30 GiB8±37%
Behemoth-X-123B-v2Q4_K_S123B64.79 GiB6.19 GiB71.68 GiB0.32 GiB5±8.3%
Mistral-Large-Instruct-2411Q4_K_S123B64.79 GiB6.19 GiB71.68 GiB0.32 GiB5±8.3%
OYM-Qimi-122B-A10B-K2.6MoEI1-Q4_K_M125B70.64 GiB0.42 GiB71.63 GiB0.37 GiB25±37%
Qwopus3.5-122B-A10B-Kimi-K2.6-destill-healed-abliteratedMoEQ4_K_M123B70.63 GiB0.42 GiB71.63 GiB0.37 GiB25±37%
Llama-3_1-Nemotron-51B-InstructIQ4_XS51.5B25.83 GiB45.00 GiB71.52 GiB0.48 GiB5±8.3%
Step-3.5-Flash-REAP-121B-A11BI1-Q4_K_S121B63.83 GiB7.04 GiB71.45 GiB0.55 GiB5±8.3%
Step-3.7-FlashIQ2_M201B63.68 GiB7.04 GiB71.30 GiB0.70 GiB5±8.3%
Mistral-Medium-3.5-128BIQ4_XS128B64.39 GiB6.19 GiB71.29 GiB0.71 GiB5±8.3%
MiMo-V2-FlashMoEKV unresolvedIQ2_XXS310B68.47 GiB2.11 GiB71.18 GiB0.82 GiB21±37%
CalmeRys-78B-Orpo-v0.1Q6_K78.0B64.27 GiB6.05 GiB71.00 GiB1.00 GiB5±8.3%
calme-2.3-rys-78bQ6_K78.0B64.27 GiB6.05 GiB71.00 GiB1.00 GiB5±8.3%
Llama-3_3-Nemotron-Super-49B-v1_5IQ4_XS49.9B25.06 GiB45.00 GiB70.75 GiB1.25 GiB5±8.3%
Llama-3_3-Nemotron-Super-49B-v1IQ4_XS49.9B25.06 GiB45.00 GiB70.75 GiB1.25 GiB5±8.3%
Valkyrie-49B-v2.1I1-IQ4_XS49.9B25.03 GiB45.00 GiB70.72 GiB1.28 GiB5±8.3%
GLM-4.6VMoEQ4_K_L108B66.89 GiB3.23 GiB70.70 GiB1.30 GiB16±37%
Qwen3.5-122B-A10BMoEQ4_K_S125B69.66 GiB0.42 GiB70.66 GiB1.34 GiB26±37%
Ornith-1.0-35B-AEON-Ultimate-Uncensored-BF16MoEQ8_035.1B69.57 GiB0.35 GiB70.48 GiB1.52 GiB26±37%
Laguna-S-2.1MoEQ4_1118B68.96 GiB0.88 GiB70.41 GiB1.59 GiB23±37%
MiniMax-M2.7MoEUD-IQ2_M229B65.32 GiB4.36 GiB70.22 GiB1.78 GiB18±37%
HarmonicHarlequin_v5-20BQ8_033.3B32.97 GiB36.56 GiB70.13 GiB1.87 GiB5±8.3%
Qwen3.5-122B-A10B-hereticMoEI1-Q4_K_M123B69.11 GiB0.42 GiB70.11 GiB1.89 GiB26±37%
Qwen2.5-Coder-32B-InstructQ8_032.8B64.86 GiB4.50 GiB70.01 GiB1.99 GiB5±8.3%
XORTRON-NXTXPRTXXLIQ4_XS128B63.03 GiB6.19 GiB69.93 GiB2.07 GiB5±8.3%
MiMo-V2.5MoEKV unresolvedIQ1_M311B67.01 GiB2.11 GiB69.72 GiB2.28 GiB22±37%
GLM-4.7-REAP-218B-A32BMoEUD-IQ1_M218B62.63 GiB6.47 GiB69.69 GiB2.31 GiB12±37%
Mistral-Small-4-119B-2603MoEUD-Q4_K_M119B68.70 GiB0.40 GiB69.67 GiB2.33 GiB26±37%
gpt-oss-120b-Uncensored-xCloudMoEI1-Q4_1117B68.42 GiB0.64 GiB69.60 GiB2.40 GiB25±37%
gpt-oss-120b-abliteratedMoEI1-Q4_1117B68.42 GiB0.64 GiB69.60 GiB2.40 GiB25±37%
HunyuanImage-2.1Q6_K17.5B68.97 GiB0.00 GiB69.57 GiB2.43 GiB5±8.3%
GLM-4.5VMoEI1-Q4_K_M108B65.61 GiB3.23 GiB69.42 GiB2.58 GiB16±37%
Devstral-2-123B-Instruct-2512IQ4_XS125B62.51 GiB6.19 GiB69.41 GiB2.59 GiB5±8.3%
Qwen3-235B-A22B-abliteratedMoEI1-IQ2_S235B65.40 GiB3.30 GiB69.29 GiB2.71 GiB16±37%
grok-2MoEIQ2_XXS270B63.81 GiB4.50 GiB69.00 GiB3.00 GiB8±37%
Qwen3-VL-235B-A22B-ThinkingMoEUD-IQ1_M236B64.90 GiB3.30 GiB68.79 GiB3.21 GiB16±37%
Qwen3-VL-235B-A22B-InstructMoEUD-IQ1_M236B64.83 GiB3.30 GiB68.72 GiB3.28 GiB16±37%
MiniMax-M2.5MoEUD-IQ1_M229B63.74 GiB4.36 GiB68.63 GiB3.37 GiB18±37%
GLM-4.5-Air-DerestrictedMoEQ4_1110B64.77 GiB3.23 GiB68.58 GiB3.42 GiB17±37%
GLM-4.5-AirMoEQ4_1110B64.77 GiB3.23 GiB68.58 GiB3.42 GiB17±37%
gpt-oss-20b-hereticMoEIQ4_NL20.9B67.58 GiB0.43 GiB68.54 GiB3.46 GiB15±37%
Llama-4-Scout-17B-16E-InstructMoEKV unresolvedQ4_1109B64.35 GiB3.38 GiB68.30 GiB3.70 GiB16±37%
MiniMax-M2.7-BF16-ultra-uncensored-hereticMoEI1-IQ2_S229B63.36 GiB4.36 GiB68.26 GiB3.74 GiB18±37%
NVIDIA-Nemotron-3-Super-120B-A12B-BF16MoEQ4_0124B66.12 GiB1.55 GiB68.21 GiB3.79 GiB21±37%
Qwen3.5-88BMoEI1-Q6_K87.7B67.08 GiB0.42 GiB68.08 GiB3.92 GiB24±37%
Mixtral-8x22B-Instruct-v0.1MoEQ3_K_M141B63.14 GiB3.94 GiB67.69 GiB4.31 GiB8±37%
Mixtral-8x22B-v0.1MoEQ3_K_M141B63.14 GiB3.94 GiB67.69 GiB4.31 GiB8±37%
Mixtral-8x22B-v0.1MoEQ3_K_M141B63.13 GiB3.94 GiB67.68 GiB4.32 GiB8±37%
GLM-4.6-REAP-268B-A32BMoEUD-TQ1_0269B60.36 GiB6.47 GiB67.42 GiB4.58 GiB13±37%
step-3.5-flashIQ2_M199B59.59 GiB7.04 GiB67.20 GiB4.80 GiB5±8.3%
Qwen3.6-35B-A3B-uncensored-heretic-Native-MTP-PreservedMoEBF1635.1B66.19 GiB0.35 GiB67.09 GiB4.91 GiB27±37%
Qwen-AgentWorld-35B-A3BMoEBF1634.7B66.19 GiB0.35 GiB67.09 GiB4.91 GiB27±37%
Qwable-v1MoEBF1636.0B66.19 GiB0.35 GiB67.09 GiB4.91 GiB27±37%
Salience-1.5-ProMoEBF1636.0B66.19 GiB0.35 GiB67.09 GiB4.91 GiB27±37%
T-SearchMoEBF1636.0B66.19 GiB0.35 GiB67.09 GiB4.91 GiB27±37%
Qwen35B-Agent-R2MoEF1634.7B66.19 GiB0.35 GiB67.09 GiB4.91 GiB27±37%
Ornith-1.0-35B-Heretic-MTPMoEBF16—66.19 GiB0.35 GiB67.09 GiB4.91 GiB27±37%
Ornith-1.0-35BMoEBF1634.7B66.19 GiB0.35 GiB67.09 GiB4.91 GiB27±37%
Carnice-Qwen3.6-MoE-35B-A3BMoEF1636.0B66.19 GiB0.35 GiB67.09 GiB4.91 GiB27±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 M2 Max run?
2071 of 2118 indexed open-weight models fit a Apple M2 Max at 65,536 context with q4_0 KV cache, the largest being MiniMax-M2.1 at IQ2_M. That covers text, vision-language, image, video and speech models.
How much usable memory does a Apple M2 Max actually have?
Its nameplate is 96 GB, but about 66.96 GiB is available to a model once driver and compositor overhead is accounted for, and only 72 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.