What AI models can a Apple M3 run?

1802 of 2118 indexed open-weight models fit a Apple M3 at 8,192 context with f16 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 M3 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 M3 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 M3

Apple M3 has 16 GB of unified memory at 102 GB/s — about 11.16 GiB usable after driver and compositor overhead. 1802 of 2118 indexed models fit at 8K context with f16 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 1548vision language 151audio asr 39image 2audio tts 21embedding 26

What fits at 8K context

largest quantization that fits, per model · 1802 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%
GLM-Z1-Rumination-32B-0414	IQ2_XS	33.1B	9.45 GiB	1.91 GiB	11.99 GiB	0.01 GiB	7±8.3%
Dolphin3.0-R1-Mistral-24B	IQ3_M	23.6B	10.07 GiB	1.25 GiB	11.99 GiB	0.01 GiB	7±8.3%
Goetia-26B-A4B-v1.4MoE	I1-IQ3_XXS	26.0B	10.84 GiB	0.61 GiB	11.99 GiB	0.01 GiB	7±8.3%
G4-Moonlight-Dusk-26B-A4B-hereticMoE	I1-IQ3_XXS	26.5B	10.84 GiB	0.61 GiB	11.99 GiB	0.01 GiB	7±8.3%
Pantheon-Reasoning-26B-A4B-1.1-hereticMoE	I1-IQ3_XXS	26.5B	10.84 GiB	0.61 GiB	11.99 GiB	0.01 GiB	7±8.3%
G4-Moonlight-Dusk-26B-A4BMoE	I1-IQ3_XXS	26.5B	10.84 GiB	0.61 GiB	11.99 GiB	0.01 GiB	7±8.3%
Chimera-X-26B-A4BMoE	I1-IQ3_XXS	26.5B	10.84 GiB	0.61 GiB	11.99 GiB	0.01 GiB	7±8.3%
Pantheon-Reasoning-26B-A4B-1.1MoE	I1-IQ3_XXS	26.5B	10.84 GiB	0.61 GiB	11.99 GiB	0.01 GiB	7±8.3%
Gemma-4-26B-A4B-StyleTune-V2MoE	I1-IQ3_XXS	26.5B	10.84 GiB	0.61 GiB	11.99 GiB	0.01 GiB	7±8.3%
Gemma-4-26B-A4B-StyleTuneMoE	I1-IQ3_XXS	26.5B	10.84 GiB	0.61 GiB	11.99 GiB	0.01 GiB	7±8.3%
gemma-4-26b-a4b-heretic-styletune-v2-headMoE	I1-IQ3_XXS	25.8B	10.84 GiB	0.61 GiB	11.99 GiB	0.01 GiB	7±8.3%
gemma-2-27b-it	Q2_K_S	27.2B	9.06 GiB	2.25 GiB	11.98 GiB	0.02 GiB	7±8.3%
OpenAI-gpt-oss-20B-Claude-4.5-Opus-Heretic-UncensoredMoE	I1-IQ2_M	20.9B	11.24 GiB	0.21 GiB	11.98 GiB	0.02 GiB	20±37%
gpt-oss-20b-uncensoredMoE	I1-IQ2_M	20.9B	11.24 GiB	0.21 GiB	11.98 GiB	0.02 GiB	20±37%
gpt-oss-safeguard-20bMoE	I1-IQ2_M	21.5B	11.24 GiB	0.21 GiB	11.98 GiB	0.02 GiB	20±37%
gpt-oss-20b-DerestrictedMoE	Q2_K	20.9B	11.24 GiB	0.21 GiB	11.98 GiB	0.02 GiB	20±37%
Huihui-gpt-oss-20b-BF16-abliterated-v2MoE	I1-IQ2_M	20.9B	11.24 GiB	0.21 GiB	11.98 GiB	0.02 GiB	20±37%
metatune-gpt20b-R1.09MoE	I1-IQ2_M	21.5B	11.24 GiB	0.21 GiB	11.98 GiB	0.02 GiB	20±37%
Magistral-Small-2509-Vision	Q2_K_L	24.0B	10.06 GiB	1.25 GiB	11.98 GiB	0.02 GiB	7±8.3%
Goetia-26B-A4B-v1.3-Absolute-Heretic-ARAMoE	I1-IQ3_XS	25.8B	10.84 GiB	0.61 GiB	11.98 GiB	0.02 GiB	7±8.3%
Frank-26B-A4BMoE	I1-IQ3_XS	26.5B	10.84 GiB	0.61 GiB	11.98 GiB	0.02 GiB	7±8.3%
G4-MeroMero-26B-A4B-it-uncensored-hereticMoE	I1-IQ3_XS	25.8B	10.84 GiB	0.61 GiB	11.98 GiB	0.02 GiB	7±8.3%
EVE-26b-XENO-HATMoE	I1-IQ3_XS	25.8B	10.84 GiB	0.61 GiB	11.98 GiB	0.02 GiB	7±8.3%
Gemma-4-26B-A4B-Animus-V14.1-FFT-hereticMoE	I1-IQ3_XS	25.8B	10.84 GiB	0.61 GiB	11.98 GiB	0.02 GiB	7±8.3%
gemma-4-26B-A4B-it-qat-q4_0-unquantized-hereticMoE	I1-IQ3_XS	25.8B	10.84 GiB	0.61 GiB	11.98 GiB	0.02 GiB	7±8.3%
Huihui-gemma-4-26B-A4B-it-qat-q4_0-unquantized-abliteratedMoE	I1-IQ3_XS	26.5B	10.84 GiB	0.61 GiB	11.98 GiB	0.02 GiB	7±8.3%
G4-MeroMero-26B-A4BMoE	I1-IQ3_XS	25.8B	10.84 GiB	0.61 GiB	11.98 GiB	0.02 GiB	7±8.3%
G4-Dark-Soul-26B-A4BMoE	I1-IQ3_XS	25.8B	10.84 GiB	0.61 GiB	11.98 GiB	0.02 GiB	7±8.3%
gemma-4-26B-A4B-it-local-abliterated-sota-internal-t34MoE	I1-IQ3_XS	25.8B	10.84 GiB	0.61 GiB	11.98 GiB	0.02 GiB	7±8.3%
gemma-4-26B-A4B-it-SOMPOA-heresyMoE	I1-IQ3_XS	25.8B	10.84 GiB	0.61 GiB	11.98 GiB	0.02 GiB	7±8.3%
gemma-4-26B-A4B-it-hereticMoE	I1-IQ3_XS	25.8B	10.84 GiB	0.61 GiB	11.98 GiB	0.02 GiB	7±8.3%
gemma-4-26B-A4B-it-abliterixMoE	I1-IQ3_XS	25.8B	10.84 GiB	0.61 GiB	11.98 GiB	0.02 GiB	7±8.3%
gemma-4-26B-A4B-it-heretic-ara-v2MoE	I1-IQ3_XS	25.8B	10.84 GiB	0.61 GiB	11.98 GiB	0.02 GiB	7±8.3%
Gemma-4-26B-A4B-it-heretic-antislopMoE	I1-IQ3_XS	25.8B	10.84 GiB	0.61 GiB	11.98 GiB	0.02 GiB	7±8.3%
gemma-4-26B-A4B-it-Claude-Opus-Distill-v2MoE	I1-IQ3_XS	26.5B	10.84 GiB	0.61 GiB	11.98 GiB	0.02 GiB	7±8.3%
gemma-4-26B-A4B-Heretic-StableMoE	I1-IQ3_XS	25.8B	10.84 GiB	0.61 GiB	11.98 GiB	0.02 GiB	7±8.3%
gemma-4-26B-A4B-it-Uncensored-MAXMoE	I1-IQ3_XS	25.8B	10.84 GiB	0.61 GiB	11.98 GiB	0.02 GiB	7±8.3%
gemma-4-26B-A4B-it-ultra-uncensored-hereticMoE	I1-IQ3_XS	25.8B	10.84 GiB	0.61 GiB	11.98 GiB	0.02 GiB	7±8.3%
gemma-4-26B-A4B-it-ara-abliteratedMoE	I1-IQ3_XS	25.8B	10.84 GiB	0.61 GiB	11.98 GiB	0.02 GiB	7±8.3%
Huihui-gemma-4-26B-A4B-it-abliteratedMoE	I1-IQ3_XS	26.5B	10.84 GiB	0.61 GiB	11.98 GiB	0.02 GiB	7±8.3%
Gemma-4-26B-A4B-AbliteratedMoE	I1-IQ3_XS	25.8B	10.84 GiB	0.61 GiB	11.98 GiB	0.02 GiB	7±8.3%
gemma4-26b-fiction-bf16MoE	I1-IQ3_XS	25.8B	10.84 GiB	0.61 GiB	11.98 GiB	0.02 GiB	7±8.3%
gemma-4-26B-A4B-it-heretic-araMoE	I1-IQ3_XS	25.8B	10.84 GiB	0.61 GiB	11.98 GiB	0.02 GiB	7±8.3%
ERNIE-21B-A3B-Claude-4.5-High-OPUS-Thinking	IQ4_XS	21.8B	10.97 GiB	0.44 GiB	11.98 GiB	0.02 GiB	7±8.3%
NVIDIA-Nemotron-Nano-12B-v2	Q6_K	12.3B	9.42 GiB	1.94 GiB	11.98 GiB	0.02 GiB	7±8.3%
InternVL3_5-30B-A3B	IQ3_XXS	30.8B	11.38 GiB	0.00 GiB	11.97 GiB	0.03 GiB	7±8.3%
Llama-3.2-8X3B-MOE-Dark-Champion-Instruct-uncensored-abliterated-18.4BMoE	Q4_K_M	18.4B	10.54 GiB	0.88 GiB	11.97 GiB	0.03 GiB	12±37%
spoomplesmaxx-v2.1-30B	I1-Q2_K_S	28.9B	9.31 GiB	2.00 GiB	11.97 GiB	0.03 GiB	7±8.3%
Huihui-granite-4.1-30b-abliterated	I1-Q2_K_S	28.9B	9.31 GiB	2.00 GiB	11.97 GiB	0.03 GiB	7±8.3%
granite-4.1-30b-heretic	I1-Q2_K_S	28.9B	9.31 GiB	2.00 GiB	11.97 GiB	0.03 GiB	7±8.3%
c4ai-command-r-08-2024	IQ2_S	32.3B	10.05 GiB	1.25 GiB	11.97 GiB	0.03 GiB	7±8.3%
medgemma-27b-it	Q2_K_L	28.8B	10.10 GiB	1.23 GiB	11.97 GiB	0.03 GiB	7±8.3%
gemma-3-27b-it-abliterated	Q2_K_L	27.4B	10.10 GiB	1.23 GiB	11.97 GiB	0.03 GiB	7±8.3%
gemma-3-27b-it	Q2_K_L	27.4B	10.10 GiB	1.23 GiB	11.97 GiB	0.03 GiB	7±8.3%
gpt-oss-20bMoE	Q6_K	21.5B	11.21 GiB	0.21 GiB	11.96 GiB	0.04 GiB	20±37%
Huihui-gpt-oss-20b-BF16-abliteratedMoE	Q6_K	20.9B	11.21 GiB	0.21 GiB	11.96 GiB	0.04 GiB	20±37%
North-Mini-Code-1.0MoE	UD-IQ3_XXS	30.5B	10.90 GiB	0.52 GiB	11.95 GiB	0.05 GiB	24±37%
Darwin-35B-A3B-OpusMoE	IQ2_M	36.0B	11.24 GiB	0.16 GiB	11.95 GiB	0.05 GiB	34±37%
Aurora-Code-1MoE	IQ2_M	34.7B	11.24 GiB	0.16 GiB	11.95 GiB	0.05 GiB	34±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 M3 run?: 1802 of 2118 indexed open-weight models fit a Apple M3 at 8,192 context with f16 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 M3 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 M3 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.