Apple · apple

Apple M4

Apple M4 has 24 GB of unified memory at 120 GB/s — about 16.74 GiB usable after driver and compositor overhead. 1927 of 2118 indexed models fit at 32K context with q4_0 KV. Note only 18 GB of its 24 GB is allocatable to the GPU.

Spec sheet· bandwidth, theoreticalFrom the file· fit from summed bytesPredicted· speed
Memory
24 GB
LPDDR5X-7500
Bandwidth
120 GB/s
128-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
vision language 171text 1652video 16image 2audio asr 39audio tts 21embedding 26

What fits at 32K context

largest quantization that fits, per model · 1927 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
Qwen3.6-35B-A3BMoEUD-IQ4_NL36.0B17.26 GiB0.18 GiB17.99 GiB0.01 GiB28±37%
Qwen3.5-88BMoEI1-IQ1_S87.7B17.20 GiB0.21 GiB17.99 GiB0.01 GiB25±37%
Qwen3-14B-GPT-5.2-High-Reasoning-DistillIQ4_NL14.8B15.96 GiB1.41 GiB17.98 GiB0.02 GiB6±8.3%
Pantheon-Reasoning-27BQ4_127.8B16.80 GiB0.56 GiB17.97 GiB0.03 GiB6±8.3%
Qwen3.5-27BQ4_127.8B16.80 GiB0.56 GiB17.97 GiB0.03 GiB6±8.3%
Wan2.1-VACE-14BQ8_017.3B17.38 GiB0.00 GiB17.97 GiB0.03 GiB6±8.3%
ERNIE-4.5-21B-A3B-ThinkingQ6_K_L21.8B16.90 GiB0.49 GiB17.96 GiB0.04 GiB6±8.3%
ERNIE-4.5-21B-A3B-PTQ6_K_L21.9B16.90 GiB0.49 GiB17.96 GiB0.04 GiB6±8.3%
Qwen3-42B-A3B-2507-Thinking-Abliterated-uncensored-TOTAL-RECALL-v2-Medium-MASTER-CODERMoEI1-IQ3_XS42.4B16.24 GiB1.18 GiB17.96 GiB0.04 GiB17±37%
xLAM-8x7b-rMoEQ2_K_L46.7B16.24 GiB1.13 GiB17.95 GiB0.05 GiB9±37%
Gemma-4-Gembrain-X-Core-31BI1-IQ4_XS31.3B15.59 GiB1.74 GiB17.95 GiB0.05 GiB6±8.3%
Gemma-4-Gembrain-X-31BI1-IQ4_XS31.3B15.59 GiB1.74 GiB17.95 GiB0.05 GiB6±8.3%
Gemma-4-31B-Isometry-Fabled-PersonaI1-IQ4_XS31.3B15.59 GiB1.74 GiB17.95 GiB0.05 GiB6±8.3%
Versipellis-31BI1-IQ4_XS31.3B15.59 GiB1.74 GiB17.95 GiB0.05 GiB6±8.3%
Gemma4-Gutenberg-31BI1-IQ4_XS31.3B15.59 GiB1.74 GiB17.95 GiB0.05 GiB6±8.3%
G4-MeroMero-31B-uncensored-hereticI1-IQ4_XS31.3B15.59 GiB1.74 GiB17.95 GiB0.05 GiB6±8.3%
Gemma-4-Novelist-31BI1-IQ4_XS31.3B15.59 GiB1.74 GiB17.95 GiB0.05 GiB6±8.3%
Wanabi-Gemma4-31BI1-IQ4_XS31.3B15.59 GiB1.74 GiB17.95 GiB0.05 GiB6±8.3%
G4-Alice-v1.2-31BI1-IQ4_XS31.3B15.59 GiB1.74 GiB17.95 GiB0.05 GiB6±8.3%
Agares-31B-v1I1-IQ4_XS30.7B15.59 GiB1.74 GiB17.95 GiB0.05 GiB6±8.3%
Gemma4-Gutenberg-31B-HereticI1-IQ4_XS31.3B15.59 GiB1.74 GiB17.95 GiB0.05 GiB6±8.3%
gemma-4-Ortenzya-The-Creative-Wordsmith-31B-it-uncensored-hereticI1-IQ4_XS31.3B15.59 GiB1.74 GiB17.95 GiB0.05 GiB6±8.3%
Gemma-4-Gemsicle-31BI1-IQ4_XS31.3B15.59 GiB1.74 GiB17.95 GiB0.05 GiB6±8.3%
Gemma-4-Gembrain-31B-it-uncensored-hereticI1-IQ4_XS31.3B15.59 GiB1.74 GiB17.95 GiB0.05 GiB6±8.3%
Melinoe-Gemma4-31B-VL-hereticI1-IQ4_XS31.3B15.59 GiB1.74 GiB17.95 GiB0.05 GiB6±8.3%
G4-MeroMero-31BI1-IQ4_XS31.3B15.59 GiB1.74 GiB17.95 GiB0.05 GiB6±8.3%
Glistening-Gem-31B-v1.0I1-IQ4_XS31.3B15.59 GiB1.74 GiB17.95 GiB0.05 GiB6±8.3%
Melinoe-Gemma4-31B-VLI1-IQ4_XS31.3B15.59 GiB1.74 GiB17.95 GiB0.05 GiB6±8.3%
Gemma-4-31B-Storymaxxed3I1-IQ4_XS31.3B15.59 GiB1.74 GiB17.95 GiB0.05 GiB6±8.3%
Huihui-gemma-4-31B-it-qat-q4_0-unquantized-abliteratedI1-IQ4_XS32.7B15.59 GiB1.74 GiB17.95 GiB0.05 GiB6±8.3%
gemma-4-31B-Queen-it-qat-q4_0-unquantizedI1-IQ4_XS31.3B15.59 GiB1.74 GiB17.95 GiB0.05 GiB6±8.3%
gemma-4-31B-it-qat-q4_0-unquantized-hereticI1-IQ4_XS31.3B15.59 GiB1.74 GiB17.95 GiB0.05 GiB6±8.3%
Gemma-4-AssGuard-31BI1-IQ4_XS31.3B15.59 GiB1.74 GiB17.95 GiB0.05 GiB6±8.3%
copywriter-gemma4-31bI1-IQ4_XS32.7B15.59 GiB1.74 GiB17.95 GiB0.05 GiB6±8.3%
gemma-4-31B-heretic-finetuneI1-IQ4_XS30.7B15.59 GiB1.74 GiB17.95 GiB0.05 GiB6±8.3%
Gemma-4-Garnet-V2-31B-it-ultra-uncensored-hereticI1-IQ4_XS31.3B15.59 GiB1.74 GiB17.95 GiB0.05 GiB6±8.3%
gemma-4-31B-it-abliterated-v3I1-IQ4_XS31.3B15.59 GiB1.74 GiB17.95 GiB0.05 GiB6±8.3%
gemma-4-31B-it-noloopI1-IQ4_XS31.3B15.59 GiB1.74 GiB17.95 GiB0.05 GiB6±8.3%
Webs-Sejong-31B-v7I1-IQ4_XS31.3B15.59 GiB1.74 GiB17.95 GiB0.05 GiB6±8.3%
Lilith-31B-v1.0I1-IQ4_XS31.3B15.59 GiB1.74 GiB17.95 GiB0.05 GiB6±8.3%
JGOS-31B-ThinkI1-IQ4_XS31.3B15.59 GiB1.74 GiB17.95 GiB0.05 GiB6±8.3%
gemma-4-31B-MergemaxxedI1-IQ4_XS31.3B15.59 GiB1.74 GiB17.95 GiB0.05 GiB6±8.3%
K1-v6-zeroI1-IQ4_XS32.7B15.59 GiB1.74 GiB17.95 GiB0.05 GiB6±8.3%
Gemma-4-Queen-31B-it-uncensored-hereticI1-IQ4_XS31.3B15.59 GiB1.74 GiB17.95 GiB0.05 GiB6±8.3%
Gemma-4-Sphinsikus-Chronist-31BI1-IQ4_XS31.3B15.59 GiB1.74 GiB17.95 GiB0.05 GiB6±8.3%
gemma-4-31B-it-hereticI1-IQ4_XS31.3B15.59 GiB1.74 GiB17.95 GiB0.05 GiB6±8.3%
Gemma4-31B-Finetuned-V2I1-IQ4_XS32.7B15.59 GiB1.74 GiB17.95 GiB0.05 GiB6±8.3%
Gemma-4-31B-storymaxxedI1-IQ4_XS31.3B15.59 GiB1.74 GiB17.95 GiB0.05 GiB6±8.3%
Gemma-4-31B-storymaxxed2I1-IQ4_XS31.3B15.59 GiB1.74 GiB17.95 GiB0.05 GiB6±8.3%
gemma-4-31B-it-Grand-Horror-X-INTENSE-HERETIC-UNCENSORED-ThinkingI1-IQ4_XS31.3B15.59 GiB1.74 GiB17.95 GiB0.05 GiB6±8.3%
gemma-4-31B-it-Mystery-Fine-Tune-HERETIC-UNCENSORED-ThinkingI1-IQ4_XS31.3B15.59 GiB1.74 GiB17.95 GiB0.05 GiB6±8.3%
gemma-4-31B-it-The-DECKARD-HERETIC-UNCENSORED-ThinkingI1-IQ4_XS31.3B15.59 GiB1.74 GiB17.95 GiB0.05 GiB6±8.3%
Huihui-gemma-4-31B-it-abliterated-v2I1-IQ4_XS32.7B15.59 GiB1.74 GiB17.95 GiB0.05 GiB6±8.3%
Gemma-4-Queen-31B-itI1-IQ4_XS31.3B15.59 GiB1.74 GiB17.95 GiB0.05 GiB6±8.3%
gemma-4-31B-it-abliteratedI1-IQ4_XS31.3B15.59 GiB1.74 GiB17.95 GiB0.05 GiB6±8.3%
gemma-4-31b-it-heretic-araI1-IQ4_XS31.3B15.59 GiB1.74 GiB17.95 GiB0.05 GiB6±8.3%
Monika-31BI1-IQ4_XS31.3B15.59 GiB1.74 GiB17.95 GiB0.05 GiB6±8.3%
Gemma-4-31B-Fable-CoderI1-IQ4_XS32.7B15.59 GiB1.74 GiB17.95 GiB0.05 GiB6±8.3%
gemma-4-31B-anthologyI1-IQ4_XS31.3B15.59 GiB1.74 GiB17.95 GiB0.05 GiB6±8.3%
Omni-31B-Turkish-Reasoning-ModelI1-IQ4_XS31.3B15.59 GiB1.74 GiB17.95 GiB0.05 GiB6±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 M4 run?
1927 of 2118 indexed open-weight models fit a Apple M4 at 32,768 context with q4_0 KV cache, the largest being Qwen3.6-35B-A3B at UD-IQ4_NL. That covers text, vision-language, image, video and speech models.
How much usable memory does a Apple M4 actually have?
Its nameplate is 24 GB, but about 16.74 GiB is available to a model once driver and compositor overhead is accounted for, and only 18 GB of the pool can be allocated to the GPU at all.
Is a Apple M4 fast for local AI?
Its memory bandwidth is 120 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 M4 — what AI models can it run locally? — ossmodeldb