What AI models can a Apple M2 run?

1820 of 2118 indexed open-weight models fit a Apple M2 at 4,096 context with q4_0 KV cache, the largest being Wan2.1-VACE-14B at Q5_K_S. That covers text, vision-language, image, video and speech models.

How much usable memory does a Apple M2 actually have?

Its nameplate is 16 GB, but about 11.16 GiB is available to a model once driver and compositor overhead is accounted for, and only 12 GB of the pool can be allocated to the GPU at all.

Is a Apple M2 fast for local AI?

Its memory bandwidth is 102 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

Apple M2 has 16 GB of unified memory at 102 GB/s — about 11.16 GiB usable after driver and compositor overhead. 1820 of 2118 indexed models fit at 4K context with q4_0 KV. Note only 12 GB of its 16 GB is allocatable to the GPU.

Spec sheet· bandwidth, theoreticalFrom the file· fit from summed bytesPredicted· speed

Memory

16 GB

LPDDR5-6400

Bandwidth

102 GB/s

128-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

video 15text 1566vision language 151audio asr 39audio tts 21image 2embedding 26

What fits at 4K context

largest quantization that fits, per model · 1820 of 2118 indexed

Model	Best quant	Params	Weights●	KV●	Total in memory◐	Headroom◐	tok/s◐
Wan2.1-VACE-14B	Q5_K_S	17.3B	11.41 GiB	0.00 GiB	12.00 GiB	0.00 GiB	7±8.3%
Trinity-Nano-PreviewMoE	BF16	6.1B	11.42 GiB	0.04 GiB	11.99 GiB	0.01 GiB	29±37%
gemma-4-26B-A4B-itMoE	IQ3_XXS	26.5B	11.33 GiB	0.13 GiB	11.99 GiB	0.01 GiB	7±8.3%
internlm2-math-plus-20b	I1-Q4_K_M	19.9B	11.16 GiB	0.21 GiB	11.98 GiB	0.02 GiB	7±8.3%
gpt-oss-20b-uncensoredMoE	IQ4_XS	20.9B	11.40 GiB	0.03 GiB	11.98 GiB	0.02 GiB	21±37%
gpt-oss-20b-hereticMoE	IQ4_XS	20.9B	11.40 GiB	0.03 GiB	11.98 GiB	0.02 GiB	21±37%
metatune-gpt20b-R1.09MoE	IQ4_XS	21.5B	11.40 GiB	0.03 GiB	11.98 GiB	0.02 GiB	21±37%
gpt-oss-20b-DerestrictedMoE	IQ4_XS	20.9B	11.40 GiB	0.03 GiB	11.98 GiB	0.02 GiB	21±37%
InternVL3_5-30B-A3B	IQ3_XXS	30.8B	11.38 GiB	0.00 GiB	11.97 GiB	0.03 GiB	7±8.3%
Trinity-2-Codestral-22B-v0.2	IQ4_XS	22.2B	11.12 GiB	0.25 GiB	11.97 GiB	0.03 GiB	7±8.3%
Cydonia-v1.3-Magnum-v4-22B	I1-IQ4_XS	22.2B	11.12 GiB	0.25 GiB	11.97 GiB	0.03 GiB	7±8.3%
Mistral-Small-22B-ArliAI-RPMax-v1.1	I1-IQ4_XS	22.2B	11.12 GiB	0.25 GiB	11.97 GiB	0.03 GiB	7±8.3%
Mistral-Small-Drummer-22B	IQ4_XS	22.2B	11.12 GiB	0.25 GiB	11.97 GiB	0.03 GiB	7±8.3%
magnum-v4-22b	I1-IQ4_XS	22.2B	11.12 GiB	0.25 GiB	11.97 GiB	0.03 GiB	7±8.3%
Codestral-22B-v0.1	IQ4_XS	22.2B	11.12 GiB	0.25 GiB	11.97 GiB	0.03 GiB	7±8.3%
Codestral-22B-v0.1-hf	IQ4_XS	22.2B	11.12 GiB	0.25 GiB	11.97 GiB	0.03 GiB	7±8.3%
Qwen3.5-21B-Claude-4.6-Opus-Deckard-Heretic-Uncensored-Thinking	I1-Q4_K_S	21.3B	11.30 GiB	0.05 GiB	11.97 GiB	0.03 GiB	7±8.3%
Qwen3.6-21B-IQ-Ultra-Heretic-Uncensored-Thinking	I1-Q4_K_S	21.3B	11.30 GiB	0.05 GiB	11.97 GiB	0.03 GiB	7±8.3%
dolphin-2.9.1-mixtral-1x22bMoE	I1-IQ4_XS	22.2B	11.11 GiB	0.25 GiB	11.97 GiB	0.03 GiB	4±37%
OLMo-2-1124-13B-Instruct	Q6_K	13.7B	10.48 GiB	0.88 GiB	11.96 GiB	0.04 GiB	7±8.3%
Laguna-XS-2.1MoE	Q2_K	33.4B	11.32 GiB	0.08 GiB	11.95 GiB	0.05 GiB	36±37%
Qwen3.5-27B-Engineer-Deckard-Gemini	I1-IQ3_S	27.7B	11.26 GiB	0.07 GiB	11.94 GiB	0.06 GiB	7±8.3%
Qwen3.5-27B-HERETIC-Polaris-Advanced-Thinking-Alpha-uncensored	I1-IQ3_S	27.4B	11.26 GiB	0.07 GiB	11.94 GiB	0.06 GiB	7±8.3%
Qwen3.5-27B-Deckard-PKD-Heretic-Uncensored-Thinking	I1-IQ3_S	27.4B	11.26 GiB	0.07 GiB	11.94 GiB	0.06 GiB	7±8.3%
Huihui-Qwen3.5-27B-abliterated	I1-IQ3_S	27.8B	11.26 GiB	0.07 GiB	11.94 GiB	0.06 GiB	7±8.3%
Qwen3.5-27B-Unredacted-MAX	I1-IQ3_S	27.4B	11.26 GiB	0.07 GiB	11.94 GiB	0.06 GiB	7±8.3%
Qwen3.5-27B-heretic	I1-IQ3_S	27.4B	11.26 GiB	0.07 GiB	11.94 GiB	0.06 GiB	7±8.3%
Qwen3.5-27B-Derestricted	I1-IQ3_S	27.8B	11.26 GiB	0.07 GiB	11.94 GiB	0.06 GiB	7±8.3%
Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled	I1-IQ3_S	27.8B	11.26 GiB	0.07 GiB	11.94 GiB	0.06 GiB	7±8.3%
Yi-1.5-6B-Chat	F16	6.1B	11.29 GiB	0.07 GiB	11.94 GiB	0.06 GiB	7±8.3%
OpenAI-gpt-oss-20B-Claude-4.5-Opus-Heretic-UncensoredMoE	I1-IQ3_M	20.9B	11.36 GiB	0.03 GiB	11.93 GiB	0.07 GiB	21±37%
gpt-oss-safeguard-20bMoE	I1-IQ3_M	21.5B	11.36 GiB	0.03 GiB	11.93 GiB	0.07 GiB	21±37%
Huihui-gpt-oss-20b-BF16-abliterated-v2MoE	I1-IQ3_M	20.9B	11.36 GiB	0.03 GiB	11.93 GiB	0.07 GiB	21±37%
Wan2.1-T2V-1.3B	Q8_0	1.4B	11.37 GiB	0.00 GiB	11.93 GiB	0.07 GiB	7±8.3%
Qwen3.6-27B-Heretic2-Thinking	I1-Q3_K_S	27.4B	11.24 GiB	0.07 GiB	11.93 GiB	0.07 GiB	7±8.3%
Qwen3.6-27B-Uncensored-Aggressive	I1-Q3_K_S	27.4B	11.24 GiB	0.07 GiB	11.93 GiB	0.07 GiB	7±8.3%
Qwen-3.5-Opus-GLM-27B	I1-Q3_K_S	26.9B	11.24 GiB	0.07 GiB	11.93 GiB	0.07 GiB	7±8.3%
Qwen3.6-27B-abliterated	I1-Q3_K_S	27.4B	11.24 GiB	0.07 GiB	11.93 GiB	0.07 GiB	7±8.3%
KoQweopus-3.5-27B-experimental	I1-Q3_K_S	27.8B	11.24 GiB	0.07 GiB	11.93 GiB	0.07 GiB	7±8.3%
Webcoda-AI-27B	I1-Q3_K_S	27.4B	11.24 GiB	0.07 GiB	11.93 GiB	0.07 GiB	7±8.3%
Qwen3.5-27B-imabari-v2	I1-Q3_K_S	27.8B	11.24 GiB	0.07 GiB	11.93 GiB	0.07 GiB	7±8.3%
Qwen3.6-27B-AEON-Ultimate-Uncensored-BF16	Q3_K_S	27.4B	11.24 GiB	0.07 GiB	11.93 GiB	0.07 GiB	7±8.3%
Qwen3.5-27B-uncensored-heretic-v1	I1-Q3_K_S	27.4B	11.24 GiB	0.07 GiB	11.93 GiB	0.07 GiB	7±8.3%
Carnice-V2-27b	I1-Q3_K_S	27.4B	11.24 GiB	0.07 GiB	11.93 GiB	0.07 GiB	7±8.3%
Qwen3.5-Queen-27B	I1-Q3_K_S	27.4B	11.24 GiB	0.07 GiB	11.93 GiB	0.07 GiB	7±8.3%
GRaPE-2-Pro	I1-Q3_K_S	27.8B	11.24 GiB	0.07 GiB	11.93 GiB	0.07 GiB	7±8.3%
Huihui-Qwen3.6-27B-abliterated	Q3_K_S	27.8B	11.24 GiB	0.07 GiB	11.93 GiB	0.07 GiB	7±8.3%
Qwen3.5-27B-abliterated	Q3_K_S	26.9B	11.24 GiB	0.07 GiB	11.93 GiB	0.07 GiB	7±8.3%
ThinkingCap-Qwen3.6-27B-heretic	Q3_K_S	27.4B	11.24 GiB	0.07 GiB	11.93 GiB	0.07 GiB	7±8.3%
Qwen3.6-27B-Claude-Opus-Reasoning-Distill-v2-heretic	Q3_K_S	27.4B	11.24 GiB	0.07 GiB	11.93 GiB	0.07 GiB	7±8.3%
Darwin-28B-REASON	I1-Q3_K_S	26.9B	11.24 GiB	0.07 GiB	11.93 GiB	0.07 GiB	7±8.3%
Huihui-Qwen3.5-27B-Claude-4.6-Opus-abliterated	I1-Q3_K_S	27.8B	11.24 GiB	0.07 GiB	11.93 GiB	0.07 GiB	7±8.3%
Qwen3.5-27B-WebNovel-Writer-zh	I1-Q3_K_S	26.9B	11.24 GiB	0.07 GiB	11.93 GiB	0.07 GiB	7±8.3%
Qwen3.5-27B_Homebrew-v2	I1-Q3_K_S	27.4B	11.24 GiB	0.07 GiB	11.93 GiB	0.07 GiB	7±8.3%
reka-flash-3.1	I1-Q4_0	20.9B	11.14 GiB	0.15 GiB	11.91 GiB	0.09 GiB	7±8.3%
reka-flash-3	Q4_0	20.9B	11.14 GiB	0.15 GiB	11.91 GiB	0.09 GiB	7±8.3%
Pantheon-Reasoning-27B	Q2_K	27.8B	11.22 GiB	0.07 GiB	11.91 GiB	0.09 GiB	7±8.3%
Qwen3.5-27B	Q2_K	27.8B	11.22 GiB	0.07 GiB	11.91 GiB	0.09 GiB	7±8.3%
Wan2.2-S2V-14B	IQ4_XS	16.3B	11.32 GiB	0.00 GiB	11.90 GiB	0.10 GiB	7±8.3%
Noromaid-20b-v0.1.1	I1-IQ4_XS	20.0B	9.94 GiB	1.36 GiB	11.89 GiB	0.11 GiB	7±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 run?: 1820 of 2118 indexed open-weight models fit a Apple M2 at 4,096 context with q4_0 KV cache, the largest being Wan2.1-VACE-14B at Q5_K_S. That covers text, vision-language, image, video and speech models.
How much usable memory does a Apple M2 actually have?: Its nameplate is 16 GB, but about 11.16 GiB is available to a model once driver and compositor overhead is accounted for, and only 12 GB of the pool can be allocated to the GPU at all.
Is a Apple M2 fast for local AI?: Its memory bandwidth is 102 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.