Apple · apple

Apple M2 Pro

Apple M2 Pro has 32 GB of unified memory at 205 GB/s — about 22.32 GiB usable after driver and compositor overhead. 1882 of 2118 indexed models fit at 128K context with q4_0 KV. Note only 24 GB of its 32 GB is allocatable to the GPU.

Spec sheet· bandwidth, theoreticalFrom the file· fit from summed bytesPredicted· speed
Memory
32 GB
LPDDR5-6400
Bandwidth
205 GB/s
256-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 1605vision language 173image 2audio tts 21audio asr 39video 16embedding 26

What fits at 128K context

largest quantization that fits, per model · 1882 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
Gemma-4-Gembrain-X-Core-31BI1-Q4_K_M31.3B17.40 GiB5.95 GiB23.99 GiB0.01 GiB7±8.3%
Gemma-4-Gembrain-X-31BI1-Q4_K_M31.3B17.40 GiB5.95 GiB23.99 GiB0.01 GiB7±8.3%
Gemma-4-31B-Isometry-Fabled-PersonaI1-Q4_K_M31.3B17.40 GiB5.95 GiB23.99 GiB0.01 GiB7±8.3%
Versipellis-31BI1-Q4_K_M31.3B17.40 GiB5.95 GiB23.99 GiB0.01 GiB7±8.3%
Gemma4-Gutenberg-31BI1-Q4_K_M31.3B17.40 GiB5.95 GiB23.99 GiB0.01 GiB7±8.3%
G4-MeroMero-31B-uncensored-hereticI1-Q4_K_M31.3B17.40 GiB5.95 GiB23.99 GiB0.01 GiB7±8.3%
Gemma-4-Novelist-31BI1-Q4_K_M31.3B17.40 GiB5.95 GiB23.99 GiB0.01 GiB7±8.3%
Wanabi-Gemma4-31BI1-Q4_K_M31.3B17.40 GiB5.95 GiB23.99 GiB0.01 GiB7±8.3%
G4-Alice-v1.2-31BI1-Q4_K_M31.3B17.40 GiB5.95 GiB23.99 GiB0.01 GiB7±8.3%
Agares-31B-v1I1-Q4_K_M30.7B17.40 GiB5.95 GiB23.99 GiB0.01 GiB7±8.3%
Gemma4-Gutenberg-31B-HereticI1-Q4_K_M31.3B17.40 GiB5.95 GiB23.99 GiB0.01 GiB7±8.3%
gemma-4-Ortenzya-The-Creative-Wordsmith-31B-it-uncensored-hereticI1-Q4_K_M31.3B17.40 GiB5.95 GiB23.99 GiB0.01 GiB7±8.3%
Gemma-4-Gemsicle-31BI1-Q4_K_M31.3B17.40 GiB5.95 GiB23.99 GiB0.01 GiB7±8.3%
Gemma-4-Gembrain-31B-it-uncensored-hereticI1-Q4_K_M31.3B17.40 GiB5.95 GiB23.99 GiB0.01 GiB7±8.3%
Melinoe-Gemma4-31B-VL-hereticI1-Q4_K_M31.3B17.40 GiB5.95 GiB23.99 GiB0.01 GiB7±8.3%
G4-MeroMero-31BI1-Q4_K_M31.3B17.40 GiB5.95 GiB23.99 GiB0.01 GiB7±8.3%
Glistening-Gem-31B-v1.0I1-Q4_K_M31.3B17.40 GiB5.95 GiB23.99 GiB0.01 GiB7±8.3%
Melinoe-Gemma4-31B-VLI1-Q4_K_M31.3B17.40 GiB5.95 GiB23.99 GiB0.01 GiB7±8.3%
Gemma-4-31B-Storymaxxed3I1-Q4_K_M31.3B17.40 GiB5.95 GiB23.99 GiB0.01 GiB7±8.3%
Huihui-gemma-4-31B-it-qat-q4_0-unquantized-abliteratedI1-Q4_K_M32.7B17.40 GiB5.95 GiB23.99 GiB0.01 GiB7±8.3%
gemma-4-31B-Queen-it-qat-q4_0-unquantizedI1-Q4_K_M31.3B17.40 GiB5.95 GiB23.99 GiB0.01 GiB7±8.3%
gemma-4-31B-it-qat-q4_0-unquantized-hereticI1-Q4_K_M31.3B17.40 GiB5.95 GiB23.99 GiB0.01 GiB7±8.3%
Gemma-4-AssGuard-31BI1-Q4_K_M31.3B17.40 GiB5.95 GiB23.99 GiB0.01 GiB7±8.3%
copywriter-gemma4-31bI1-Q4_K_M32.7B17.40 GiB5.95 GiB23.99 GiB0.01 GiB7±8.3%
gemma-4-31B-heretic-finetuneI1-Q4_K_M30.7B17.40 GiB5.95 GiB23.99 GiB0.01 GiB7±8.3%
Gemma-4-Garnet-V2-31B-it-ultra-uncensored-hereticI1-Q4_K_M31.3B17.40 GiB5.95 GiB23.99 GiB0.01 GiB7±8.3%
gemma-4-31B-it-Claude-Opus-Distill-v2Q4_K_M32.7B17.40 GiB5.95 GiB23.99 GiB0.01 GiB7±8.3%
gemma-4-31B-it-abliterated-v3I1-Q4_K_M31.3B17.40 GiB5.95 GiB23.99 GiB0.01 GiB7±8.3%
Gemma-4-Harmonia-31B-uncensored-hereticQ4_K_M31.3B17.40 GiB5.95 GiB23.99 GiB0.01 GiB7±8.3%
gemma-4-31B-it-noloopI1-Q4_K_M31.3B17.40 GiB5.95 GiB23.99 GiB0.01 GiB7±8.3%
Webs-Sejong-31B-v7I1-Q4_K_M31.3B17.40 GiB5.95 GiB23.99 GiB0.01 GiB7±8.3%
Lilith-31B-v1.0I1-Q4_K_M31.3B17.40 GiB5.95 GiB23.99 GiB0.01 GiB7±8.3%
JGOS-31B-ThinkI1-Q4_K_M31.3B17.40 GiB5.95 GiB23.99 GiB0.01 GiB7±8.3%
gemma-4-31B-MergemaxxedI1-Q4_K_M31.3B17.40 GiB5.95 GiB23.99 GiB0.01 GiB7±8.3%
K1-v6-zeroI1-Q4_K_M32.7B17.40 GiB5.95 GiB23.99 GiB0.01 GiB7±8.3%
gemma-4-31B-it-uncensored-hereticQ4_K_M31.3B17.40 GiB5.95 GiB23.99 GiB0.01 GiB7±8.3%
Gemma-4-Queen-31B-it-uncensored-hereticI1-Q4_K_M31.3B17.40 GiB5.95 GiB23.99 GiB0.01 GiB7±8.3%
Gemma-4-Sphinsikus-Chronist-31BI1-Q4_K_M31.3B17.40 GiB5.95 GiB23.99 GiB0.01 GiB7±8.3%
gemma-4-31B-it-hereticI1-Q4_K_M31.3B17.40 GiB5.95 GiB23.99 GiB0.01 GiB7±8.3%
Gemma4-31B-Finetuned-V2I1-Q4_K_M32.7B17.40 GiB5.95 GiB23.99 GiB0.01 GiB7±8.3%
Gemma-4-31B-storymaxxedI1-Q4_K_M31.3B17.40 GiB5.95 GiB23.99 GiB0.01 GiB7±8.3%
Gemma-4-31B-Fable-5-Agent-DistillQ4_K_M32.7B17.40 GiB5.95 GiB23.99 GiB0.01 GiB7±8.3%
gemma-4-31B-it-uncensoredQ4_K_M31.3B17.40 GiB5.95 GiB23.99 GiB0.01 GiB7±8.3%
Gemma-4-31B-storymaxxed2I1-Q4_K_M31.3B17.40 GiB5.95 GiB23.99 GiB0.01 GiB7±8.3%
Gemma-4-Giftige-Blume-31B-v2Q4_K_M31.3B17.40 GiB5.95 GiB23.99 GiB0.01 GiB7±8.3%
gemma-4-31B-it-Grand-Horror-X-INTENSE-HERETIC-UNCENSORED-ThinkingI1-Q4_K_M31.3B17.40 GiB5.95 GiB23.99 GiB0.01 GiB7±8.3%
gemma-4-31B-it-Mystery-Fine-Tune-HERETIC-UNCENSORED-ThinkingI1-Q4_K_M31.3B17.40 GiB5.95 GiB23.99 GiB0.01 GiB7±8.3%
gemma-4-31B-it-The-DECKARD-HERETIC-UNCENSORED-ThinkingI1-Q4_K_M31.3B17.40 GiB5.95 GiB23.99 GiB0.01 GiB7±8.3%
Huihui-gemma-4-31B-it-abliterated-v2I1-Q4_K_M32.7B17.40 GiB5.95 GiB23.99 GiB0.01 GiB7±8.3%
Gemma-4-Queen-31B-itI1-Q4_K_M31.3B17.40 GiB5.95 GiB23.99 GiB0.01 GiB7±8.3%
gemma-4-31B-it-abliteratedI1-Q4_K_M31.3B17.40 GiB5.95 GiB23.99 GiB0.01 GiB7±8.3%
gemma-4-31b-it-heretic-araI1-Q4_K_M31.3B17.40 GiB5.95 GiB23.99 GiB0.01 GiB7±8.3%
Monika-31BI1-Q4_K_M31.3B17.40 GiB5.95 GiB23.99 GiB0.01 GiB7±8.3%
gemma-4-31B-it-uncensoredQ4_K_M32.7B17.40 GiB5.95 GiB23.99 GiB0.01 GiB7±8.3%
Gemma-4-31B-Fable-CoderI1-Q4_K_M32.7B17.40 GiB5.95 GiB23.99 GiB0.01 GiB7±8.3%
gemma-4-31B-anthologyI1-Q4_K_M31.3B17.40 GiB5.95 GiB23.99 GiB0.01 GiB7±8.3%
Omni-31B-Turkish-Reasoning-ModelI1-Q4_K_M31.3B17.40 GiB5.95 GiB23.99 GiB0.01 GiB7±8.3%
gemma-4-31b-kairosI1-Q4_K_M31.3B17.40 GiB5.95 GiB23.99 GiB0.01 GiB7±8.3%
gemma-4-31BI1-Q4_K_M32.7B17.40 GiB5.95 GiB23.99 GiB0.01 GiB7±8.3%
dolphin-2.7-mixtral-8x7bMoEQ3_K_S46.7B18.90 GiB4.50 GiB23.99 GiB0.01 GiB9±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 Pro run?
1882 of 2118 indexed open-weight models fit a Apple M2 Pro at 131,072 context with q4_0 KV cache, the largest being Gemma-4-Gembrain-X-Core-31B at I1-Q4_K_M. That covers text, vision-language, image, video and speech models.
How much usable memory does a Apple M2 Pro actually have?
Its nameplate is 32 GB, but about 22.32 GiB is available to a model once driver and compositor overhead is accounted for, and only 24 GB of the pool can be allocated to the GPU at all.
Is a Apple M2 Pro fast for local AI?
Its memory bandwidth is 205 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.