What AI models can a Apple M2 Max run?

2064 of 2118 indexed open-weight models fit a Apple M2 Max at 131,072 context with q4_0 KV cache, the largest being Step-3.7-Flash at UD-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.

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. 2064 of 2118 indexed models fit at 128K 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

—

Context4K 8K 16K 32K 64K 128K

KV cachef16 q8_0 q4_0quantizing the KV cache is a ~2× lever on the dominant term at long context

vision language 186text 1774image 2audio asr 39audio tts 21video 16embedding 26

What fits at 128K context

largest quantization that fits, per model · 2064 of 2118 indexed

Model	Best quant	Params	Weights●	KV●	Total in memory◐	Headroom◐	tok/s◐
Step-3.7-Flash	UD-IQ2_M	201B	57.58 GiB	13.79 GiB	71.94 GiB	0.06 GiB	5±8.3%
HuatuoGPT-o1-72B	Q6_K	72.7B	59.93 GiB	11.25 GiB	71.86 GiB	0.14 GiB	5±8.3%
Rombo-LLM-V3.0-Qwen-72b	Q6_K	72.7B	59.93 GiB	11.25 GiB	71.86 GiB	0.14 GiB	5±8.3%
Qwen2.5-72B-Instruct-abliterated	Q6_K	72.7B	59.93 GiB	11.25 GiB	71.86 GiB	0.14 GiB	5±8.3%
EVA-Qwen2.5-72B-v0.2	Q6_K	72.7B	59.93 GiB	11.25 GiB	71.86 GiB	0.14 GiB	5±8.3%
MiroThinker-v1.0-72B	Q6_K	72.7B	59.93 GiB	11.25 GiB	71.86 GiB	0.14 GiB	5±8.3%
Qwen2.5-Math-72B-Instruct	Q6_K	72.7B	59.93 GiB	11.25 GiB	71.86 GiB	0.14 GiB	5±8.3%
Qwen2.5-72B-Instruct	Q6_K	72.7B	59.93 GiB	11.25 GiB	71.86 GiB	0.14 GiB	5±8.3%
Qwen2.5-72B	Q6_K	72.7B	59.93 GiB	11.25 GiB	71.86 GiB	0.14 GiB	5±8.3%
Kimi-Dev-72B	Q6_K	72.7B	59.93 GiB	11.25 GiB	71.86 GiB	0.14 GiB	5±8.3%
magnum-v4-72b	Q6_K	72.7B	59.93 GiB	11.25 GiB	71.86 GiB	0.14 GiB	5±8.3%
KAT-Dev-72B-Exp	Q6_K	72.7B	59.93 GiB	11.25 GiB	71.86 GiB	0.14 GiB	5±8.3%
Chuluun-Qwen2.5-72B-v0.01	Q6_K	72.7B	59.93 GiB	11.25 GiB	71.86 GiB	0.14 GiB	5±8.3%
Homer-v1.0-Qwen2.5-72B	Q6_K	72.7B	59.93 GiB	11.25 GiB	71.86 GiB	0.14 GiB	5±8.3%
Qwen2.5-VL-72B-Instruct	Q6_K	73.4B	59.93 GiB	11.25 GiB	71.86 GiB	0.14 GiB	5±8.3%
Tower-Plus-72B-ultra-uncensored-heretic	I1-Q6_K	72.7B	59.93 GiB	11.25 GiB	71.86 GiB	0.14 GiB	5±8.3%
Chronos-Platinum-72B	Q6_K	72.7B	59.93 GiB	11.25 GiB	71.86 GiB	0.14 GiB	5±8.3%
UI-TARS-72B-DPO	Q6_K	73.4B	59.93 GiB	11.25 GiB	71.86 GiB	0.14 GiB	5±8.3%
MiMo-V2.5MoEKV unresolved	IQ1_M	311B	67.01 GiB	4.22 GiB	71.83 GiB	0.17 GiB	18±37%
GLM-4.5-Air-DerestrictedMoE	Q4_1	110B	64.77 GiB	6.47 GiB	71.82 GiB	0.18 GiB	13±37%
GLM-4.5-AirMoE	Q4_1	110B	64.77 GiB	6.47 GiB	71.82 GiB	0.18 GiB	13±37%
archangel_sft-kto_llama30b	IQ4_XS	32.5B	16.28 GiB	54.84 GiB	71.74 GiB	0.26 GiB	5±8.3%
Llama-4-Scout-17B-16E-InstructMoEKV unresolved	Q4_1	109B	64.35 GiB	6.75 GiB	71.68 GiB	0.32 GiB	13±37%
Mixtral-8x22B-Instruct-v0.1MoE	Q3_K_M	141B	63.14 GiB	7.88 GiB	71.62 GiB	0.38 GiB	7±37%
Mixtral-8x22B-v0.1MoE	Q3_K_M	141B	63.14 GiB	7.88 GiB	71.62 GiB	0.38 GiB	7±37%
Mixtral-8x22B-v0.1MoE	Q3_K_M	141B	63.13 GiB	7.88 GiB	71.62 GiB	0.38 GiB	7±37%
Wizard-Vicuna-30B-Uncensored	I1-IQ4_XS	32.5B	16.15 GiB	54.84 GiB	71.62 GiB	0.38 GiB	5±8.3%
MiniMax-M2.7-BF16-ultra-uncensored-hereticMoE	I1-IQ2_XS	229B	62.35 GiB	8.72 GiB	71.61 GiB	0.39 GiB	13±37%
MiniMax-M2.1MoE	I1-IQ2_XS	229B	62.35 GiB	8.72 GiB	71.61 GiB	0.39 GiB	13±37%
MiniMax-M2.5MoE	I1-IQ2_XS	229B	62.35 GiB	8.72 GiB	71.61 GiB	0.39 GiB	13±37%
WizardLM-Uncensored-SuperCOT-StoryTelling-30b	Q3_K_L	32.5B	16.09 GiB	54.84 GiB	71.56 GiB	0.44 GiB	5±8.3%
Hy3MoE	IQ1_S	299B	59.65 GiB	11.25 GiB	71.49 GiB	0.51 GiB	11±37%
Qwen3-235B-A22B-abliteratedMoE	I1-IQ2_XS	235B	64.09 GiB	6.61 GiB	71.29 GiB	0.71 GiB	13±37%
MiniMax-M2.1-REAP-139B-A10BMoE	I1-Q3_K_M	139B	62.01 GiB	8.72 GiB	71.27 GiB	0.73 GiB	12±37%
m51Lab-MiniMax-M2.7-REAP-139B-A10BMoE	I1-Q3_K_M	139B	62.01 GiB	8.72 GiB	71.27 GiB	0.73 GiB	12±37%
Laguna-S-2.1MoE	Q4_1	118B	68.96 GiB	1.73 GiB	71.26 GiB	0.74 GiB	21±37%
Qwen3.5-122B-A10BMoE	Q4_K_S	125B	69.66 GiB	0.84 GiB	71.09 GiB	0.91 GiB	24±37%
GLM-Z1-Rumination-32B-0414	BF16	33.1B	61.74 GiB	8.58 GiB	70.96 GiB	1.04 GiB	5±8.3%
Ornith-1.0-35B-AEON-Ultimate-Uncensored-BF16MoE	Q8_0	35.1B	69.57 GiB	0.70 GiB	70.83 GiB	1.17 GiB	25±37%
Caller	BF16	32.8B	61.04 GiB	9.00 GiB	70.68 GiB	1.32 GiB	5±8.3%
Dumpling-Qwen2.5-32B	BF16	32.8B	61.04 GiB	9.00 GiB	70.68 GiB	1.32 GiB	5±8.3%
OpenThinker-32B	F16	32.8B	61.04 GiB	9.00 GiB	70.68 GiB	1.32 GiB	5±8.3%
INTELLECT-2	BF16	32.8B	61.04 GiB	9.00 GiB	70.68 GiB	1.32 GiB	5±8.3%
openhands-lm-32b-v0.1	BF16	32.8B	61.04 GiB	9.00 GiB	70.68 GiB	1.32 GiB	5±8.3%
LongWriter-Zero-32B	BF16	32.8B	61.04 GiB	9.00 GiB	70.68 GiB	1.32 GiB	5±8.3%
OpenCodeReasoning-Nemotron-32B	BF16	32.8B	61.04 GiB	9.00 GiB	70.68 GiB	1.32 GiB	5±8.3%
OpenCodeReasoning-Nemotron-32B-IOI	BF16	32.8B	61.04 GiB	9.00 GiB	70.68 GiB	1.32 GiB	5±8.3%
Qwen2.5-Coder-32B-Instruct-abliterated	F16	32.8B	61.04 GiB	9.00 GiB	70.68 GiB	1.32 GiB	5±8.3%
QwQ-32B-ArliAI-RpR-v4	BF16	32.8B	61.04 GiB	9.00 GiB	70.68 GiB	1.32 GiB	5±8.3%
OpenThinker2-32B	BF16	32.8B	61.04 GiB	9.00 GiB	70.68 GiB	1.32 GiB	5±8.3%
Qwen2.5-Coder-32B	F16	32.8B	61.04 GiB	9.00 GiB	70.68 GiB	1.32 GiB	5±8.3%
Qwen2.5-32B-Instruct	F16	32.8B	61.04 GiB	9.00 GiB	70.68 GiB	1.32 GiB	5±8.3%
QwQ-32B-Preview	BF16	32.8B	61.04 GiB	9.00 GiB	70.68 GiB	1.32 GiB	5±8.3%
Qwen2.5-32b-RP-Ink	F16	32.8B	61.04 GiB	9.00 GiB	70.68 GiB	1.32 GiB	5±8.3%
deepseek-r1-qwen-2.5-32B-ablated	BF16	32.8B	61.04 GiB	9.00 GiB	70.68 GiB	1.32 GiB	5±8.3%
Rombos-LLM-V2.5-Qwen-32b	F16	32.8B	61.04 GiB	9.00 GiB	70.68 GiB	1.32 GiB	5±8.3%
DeepSeek-R1-Distill-Qwen-32B-abliterated	BF16	32.8B	61.04 GiB	9.00 GiB	70.68 GiB	1.32 GiB	5±8.3%
Qwen2.5-32B-ArliAI-RPMax-v1.3	F16	32.8B	61.04 GiB	9.00 GiB	70.68 GiB	1.32 GiB	5±8.3%
DeepSeek-R1-Distill-Qwen-32B	F16	32.8B	61.04 GiB	9.00 GiB	70.68 GiB	1.32 GiB	5±8.3%
Qwen2.5-VL-32B-Instruct	BF16	33.5B	61.04 GiB	9.00 GiB	70.68 GiB	1.32 GiB	5±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 M2 Max run?: 2064 of 2118 indexed open-weight models fit a Apple M2 Max at 131,072 context with q4_0 KV cache, the largest being Step-3.7-Flash at UD-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.