Apple · apple

Apple M4

Apple M4 has 12 GB of unified memory at 120 GB/s — about 8.37 GiB usable after driver and compositor overhead. 1417 of 2118 indexed models fit at 64K context with q4_0 KV. Note only 9 GB of its 12 GB is allocatable to the GPU.

Spec sheet· bandwidth, theoreticalFrom the file· fit from summed bytesPredicted· speed
Memory
12 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
text 1209video 12vision language 110embedding 26audio tts 21audio asr 38image 1

What fits at 64K context

largest quantization that fits, per model · 1417 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
Ling-liteMoEQ3_K_S16.8B7.47 GiB0.98 GiB9.00 GiB0.00 GiB24±37%
Le-Chaton-Slim-23BMoEI1-IQ2_S23.3B6.61 GiB1.83 GiB9.00 GiB0.00 GiB15±37%
EVA-abliterated-TIES-Qwen2.5-14BI1-Q2_K_S14.8B5.03 GiB3.38 GiB9.00 GiB0.00 GiB11±8.3%
Neuron-V1-14B-InstructI1-Q2_K_S14.8B5.03 GiB3.38 GiB9.00 GiB0.00 GiB11±8.3%
Ektome-Qwen2.5-Coder-14B-Instruct-PristinelyUncensoredI1-Q2_K_S14.8B5.03 GiB3.38 GiB9.00 GiB0.00 GiB11±8.3%
Qwen2.5-14B-Instruct-1M-abliteratedI1-Q2_K_S14.8B5.03 GiB3.38 GiB9.00 GiB0.00 GiB11±8.3%
Deepseeker-Kunou-Qwen2.5-14bI1-Q2_K_S14.8B5.03 GiB3.38 GiB9.00 GiB0.00 GiB11±8.3%
14B-Qwen2.5-Kunou-v1I1-Q2_K_S14.8B5.03 GiB3.38 GiB9.00 GiB0.00 GiB11±8.3%
Sugoi-14B-Ultra-HFI1-Q2_K_S14.8B5.03 GiB3.38 GiB9.00 GiB0.00 GiB11±8.3%
DeepSeek-R1-Distill-Qwen-14B-abliterated-v2I1-Q2_K_S14.8B5.03 GiB3.38 GiB9.00 GiB0.00 GiB11±8.3%
C1-TachuI1-Q2_K_S14.8B5.03 GiB3.38 GiB9.00 GiB0.00 GiB11±8.3%
DeepSeek-R1-Distill-Qwen-14B-abliteratedI1-Q2_K_S14.8B5.03 GiB3.38 GiB9.00 GiB0.00 GiB11±8.3%
Tessera-4I1-Q2_K_S14.8B5.03 GiB3.38 GiB9.00 GiB0.00 GiB11±8.3%
Tessera-4.1I1-Q2_K_S14.8B5.03 GiB3.38 GiB9.00 GiB0.00 GiB11±8.3%
AceReason-Nemotron-14BI1-Q2_K_S14.8B5.03 GiB3.38 GiB9.00 GiB0.00 GiB11±8.3%
UwU-14B-Math-v0.2I1-Q2_K_S14.8B5.03 GiB3.38 GiB9.00 GiB0.00 GiB11±8.3%
Impish_QWEN_14B-1MI1-Q2_K_S14.8B5.03 GiB3.38 GiB9.00 GiB0.00 GiB11±8.3%
Wan2.1-T2V-14BQ4_014.3B8.41 GiB0.00 GiB9.00 GiB0.00 GiB11±8.3%
Lamarck-14B-v0.7I1-Q2_K_S14.8B5.02 GiB3.38 GiB9.00 GiB0.00 GiB11±8.3%
QwenStock-14BI1-Q2_K_S14.8B5.02 GiB3.38 GiB9.00 GiB0.00 GiB11±8.3%
DeepSeek-R1-Distill-Qwen-14B-UncensoredI1-Q2_K_S14.8B5.02 GiB3.38 GiB9.00 GiB0.00 GiB11±8.3%
GLM-4.6V-FlashQ6_K10.3B7.70 GiB0.70 GiB8.99 GiB0.01 GiB11±8.3%
GLM-Z1-9B-0414Q6_K9.4B7.70 GiB0.70 GiB8.99 GiB0.01 GiB11±8.3%
glm4.1v-9b-base-sftI1-Q6_K10.3B7.70 GiB0.70 GiB8.99 GiB0.01 GiB11±8.3%
GLM-4-9B-0414Q6_K9.4B7.70 GiB0.70 GiB8.99 GiB0.01 GiB11±8.3%
GLM-4.1V-9B-ThinkingQ6_K10.3B7.70 GiB0.70 GiB8.99 GiB0.01 GiB11±8.3%
Grug-12BQ4_K_M12.0B7.14 GiB1.26 GiB8.99 GiB0.01 GiB11±8.3%
gemma-4-12B-it-Esper4Q4_K_M12.0B7.14 GiB1.26 GiB8.99 GiB0.01 GiB11±8.3%
gemma-4-12B-itQ4_K_M12.0B7.14 GiB1.26 GiB8.99 GiB0.01 GiB11±8.3%
Assistant_Pepe_8BQ6_K—6.15 GiB2.25 GiB8.99 GiB0.01 GiB11±8.3%
Aya-Medikal-V2I1-Q6_K8.0B6.14 GiB2.25 GiB8.99 GiB0.01 GiB11±8.3%
SuperGemma-4-12b-abliteratedI1-Q4_112.0B7.13 GiB1.26 GiB8.98 GiB0.02 GiB11±8.3%
gemma-4-12B-agentic-fable5-composer2.5-v2-3.5x-tau2-uncensored-hereticI1-Q4_112.0B7.13 GiB1.26 GiB8.98 GiB0.02 GiB11±8.3%
gemma-4-12B-coder-fable5-composer2.5-v1-uncensored-hereticI1-Q4_112.0B7.13 GiB1.26 GiB8.98 GiB0.02 GiB11±8.3%
gemma-4-12B-it-uncensored-hereticI1-Q4_112.0B7.13 GiB1.26 GiB8.98 GiB0.02 GiB11±8.3%
Aura-Medium-v1-BF16I1-Q4_112.0B7.13 GiB1.26 GiB8.98 GiB0.02 GiB11±8.3%
gemma-4-12B-it-GuardpointI1-Q4_112.0B7.13 GiB1.26 GiB8.98 GiB0.02 GiB11±8.3%
Gemma-4-12B-it-AEON-Abliterated-K4-BF16I1-Q4_112.0B7.13 GiB1.26 GiB8.98 GiB0.02 GiB11±8.3%
gemma-4-12B-it-Tachibana-AgentI1-Q4_112.0B7.13 GiB1.26 GiB8.98 GiB0.02 GiB11±8.3%
gemma-4-12b-marvin-gutenberg-rp-v2I1-Q4_112.0B7.13 GiB1.26 GiB8.98 GiB0.02 GiB11±8.3%
gemma-4-12b-crownelius-writerI1-Q4_112.0B7.13 GiB1.26 GiB8.98 GiB0.02 GiB11±8.3%
Huihui-gemma-4-12B-coder-fable5-composer2.5-v1-abliteratedI1-Q4_112.0B7.13 GiB1.26 GiB8.98 GiB0.02 GiB11±8.3%
gemma-4-12b-asterion-agenticI1-Q4_112.0B7.13 GiB1.26 GiB8.98 GiB0.02 GiB11±8.3%
Huihui-gemma-4-12B-agentic-fable5-abliteratedI1-Q4_112.0B7.13 GiB1.26 GiB8.98 GiB0.02 GiB11±8.3%
g4-12b-it-trismegistusI1-Q4_112.0B7.13 GiB1.26 GiB8.98 GiB0.02 GiB11±8.3%
gemma4-12b-it-asimovI1-Q4_112.0B7.13 GiB1.26 GiB8.98 GiB0.02 GiB11±8.3%
FabGemmaI1-Q4_112.0B7.13 GiB1.26 GiB8.98 GiB0.02 GiB11±8.3%
Huihui-gemma-4-12B-it-qat-q4_0-unquantized-abliteratedI1-Q4_112.0B7.13 GiB1.26 GiB8.98 GiB0.02 GiB11±8.3%
gemma-4-12B-it-abliterated-uncensoredI1-Q4_112.0B7.13 GiB1.26 GiB8.98 GiB0.02 GiB11±8.3%
Gemma-4-12b-it-AbliteratedI1-Q4_112.0B7.13 GiB1.26 GiB8.98 GiB0.02 GiB11±8.3%
gemma-4-12B-Queen-it-qat-q4_0-unquantizedI1-Q4_112.0B7.13 GiB1.26 GiB8.98 GiB0.02 GiB11±8.3%
gemma-4-12B-it-heretic_decensoredI1-Q4_112.0B7.13 GiB1.26 GiB8.98 GiB0.02 GiB11±8.3%
Iris-12B-gemma-4-it-qatI1-Q4_112.0B7.13 GiB1.26 GiB8.98 GiB0.02 GiB11±8.3%
gemma-4-12B-coder-fable5-composer2.5-v1I1-Q4_112.0B7.13 GiB1.26 GiB8.98 GiB0.02 GiB11±8.3%
G4-Starry-Ocean-12BI1-Q4_111.9B7.13 GiB1.26 GiB8.98 GiB0.02 GiB11±8.3%
gemma-4-12B-it-QAT-SOMPOA-heresyI1-Q4_112.0B7.13 GiB1.26 GiB8.98 GiB0.02 GiB11±8.3%
gemma-4-12B-it-uncensored-opus4.7-cotI1-Q4_112.0B7.13 GiB1.26 GiB8.98 GiB0.02 GiB11±8.3%
Gemma4-12B-IT-AbliteratedI1-Q4_112.0B7.13 GiB1.26 GiB8.98 GiB0.02 GiB11±8.3%
gemma-4-12b-it-uncensoredI1-Q4_112.0B7.13 GiB1.26 GiB8.98 GiB0.02 GiB11±8.3%
Huihui-gemma-4-12B-it-abliteratedI1-Q4_112.0B7.13 GiB1.26 GiB8.98 GiB0.02 GiB11±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?
1417 of 2118 indexed open-weight models fit a Apple M4 at 65,536 context with q4_0 KV cache, the largest being Ling-lite at Q3_K_S. That covers text, vision-language, image, video and speech models.
How much usable memory does a Apple M4 actually have?
Its nameplate is 12 GB, but about 8.37 GiB is available to a model once driver and compositor overhead is accounted for, and only 9 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.