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. 1687 of 2118 indexed models fit at 4K 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 1462video 12vision language 125audio asr 39audio tts 21embedding 26image 2

What fits at 4K context

largest quantization that fits, per model · 1687 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
SOLAR-10.7B-Instruct-v1.0-uncensoredQ6_K10.7B8.20 GiB0.21 GiB9.00 GiB0.00 GiB11±8.3%
Nous-Hermes-2-SOLAR-10.7BQ6_K10.7B8.20 GiB0.21 GiB9.00 GiB0.00 GiB11±8.3%
SOLAR-10.7B-Instruct-v1.0I1-Q6_K10.7B8.20 GiB0.21 GiB9.00 GiB0.00 GiB11±8.3%
ERNIE-21B-A3B-Thinking-Gemini-3-Pro-High-Reasoning-V2I1-IQ3_XS21.8B8.37 GiB0.06 GiB9.00 GiB0.00 GiB11±8.3%
ERNIE-21B-A3B-Claude-4.5-High-OPUS-ThinkingI1-IQ3_XS21.8B8.37 GiB0.06 GiB9.00 GiB0.00 GiB11±8.3%
ERNIE-4.5-21B-A3B-ThinkingI1-IQ3_XS21.8B8.37 GiB0.06 GiB9.00 GiB0.00 GiB11±8.3%
codegeex4-all-9bQ6_K9.4B7.69 GiB0.70 GiB9.00 GiB0.00 GiB11±8.3%
Wan2.1-T2V-14BQ4_014.3B8.41 GiB0.00 GiB9.00 GiB0.00 GiB11±8.3%
gemma-4-A4B-98e-v6-coder-itMoEIQ3_XXS20.5B8.33 GiB0.13 GiB9.00 GiB0.00 GiB11±8.3%
Qwen3-VL-8B-Instruct-HereticI1-Q3_K_L8.8B8.25 GiB0.16 GiB8.99 GiB0.01 GiB11±8.3%
glm-4-9b-chat-abliteratedQ6_K9.4B7.69 GiB0.70 GiB8.99 GiB0.01 GiB11±8.3%
glm-4-9b-chatQ6_K9.4B7.69 GiB0.70 GiB8.99 GiB0.01 GiB11±8.3%
Qwen3-Coder-30B-A3B-InstructMoEUD-IQ1_S30.5B8.34 GiB0.11 GiB8.99 GiB0.01 GiB40±37%
Apriel-1.6-15b-ThinkerI1-Q4_K_M14.9B8.18 GiB0.21 GiB8.99 GiB0.01 GiB11±8.3%
Muse-Glimmer-30BIQ2_XXS29.8B8.31 GiB0.04 GiB8.98 GiB0.02 GiB12±8.3%
internlm2-math-plus-20bI1-Q3_K_S19.9B8.16 GiB0.21 GiB8.98 GiB0.02 GiB12±8.3%
GLM-4.7-Flash-REAP-23B-A3B-absolute-heresyMoEI1-IQ3_XXS23.0B8.36 GiB0.06 GiB8.98 GiB0.02 GiB37±37%
InternVL3_5-14BQ4_K_M15.1B8.38 GiB0.00 GiB8.98 GiB0.02 GiB11±8.3%
MythoMax-L2-Kimiko-v2-13bQ4_K_M13.0B7.51 GiB0.88 GiB8.98 GiB0.02 GiB11±8.3%
MythoMax-L2-13bI1-Q4_K_M13.0B7.51 GiB0.88 GiB8.98 GiB0.02 GiB11±8.3%
spoomplesmaxx-v2.1-30BI1-IQ2_XS28.9B8.04 GiB0.28 GiB8.98 GiB0.02 GiB12±8.3%
Huihui-granite-4.1-30b-abliteratedI1-IQ2_XS28.9B8.04 GiB0.28 GiB8.98 GiB0.02 GiB12±8.3%
granite-4.1-30b-hereticI1-IQ2_XS28.9B8.04 GiB0.28 GiB8.98 GiB0.02 GiB12±8.3%
HunyuanVideo-1.5Q8_08.3B8.38 GiB0.00 GiB8.97 GiB0.03 GiB12±8.3%
rnj-1-instructQ8_08.3B8.23 GiB0.14 GiB8.97 GiB0.03 GiB12±8.3%
Grug-12BQ5_K_M12.0B8.17 GiB0.20 GiB8.97 GiB0.03 GiB12±8.3%
gemma-4-12B-it-Esper4Q5_K_M12.0B8.17 GiB0.20 GiB8.97 GiB0.03 GiB12±8.3%
gemma-4-12B-itQ5_K_M12.0B8.17 GiB0.20 GiB8.97 GiB0.03 GiB12±8.3%
Wizard-Vicuna-30B-UncensoredI1-IQ1_S32.5B6.63 GiB1.71 GiB8.97 GiB0.03 GiB12±8.3%
archangel_sft-kto_llama30bI1-IQ1_S32.5B6.63 GiB1.71 GiB8.97 GiB0.03 GiB12±8.3%
Goetia-26B-A4B-v1.4MoEI1-IQ1_M26.0B8.30 GiB0.13 GiB8.96 GiB0.04 GiB11±8.3%
G4-Moonlight-Dusk-26B-A4B-hereticMoEI1-IQ1_M26.5B8.30 GiB0.13 GiB8.96 GiB0.04 GiB11±8.3%
Pantheon-Reasoning-26B-A4B-1.1-hereticMoEI1-IQ1_M26.5B8.30 GiB0.13 GiB8.96 GiB0.04 GiB11±8.3%
G4-Moonlight-Dusk-26B-A4BMoEI1-IQ1_M26.5B8.30 GiB0.13 GiB8.96 GiB0.04 GiB11±8.3%
Chimera-X-26B-A4BMoEI1-IQ1_M26.5B8.30 GiB0.13 GiB8.96 GiB0.04 GiB11±8.3%
Pantheon-Reasoning-26B-A4B-1.1MoEI1-IQ1_M26.5B8.30 GiB0.13 GiB8.96 GiB0.04 GiB11±8.3%
Gemma-4-26B-A4B-StyleTune-V2MoEI1-IQ1_M26.5B8.30 GiB0.13 GiB8.96 GiB0.04 GiB11±8.3%
Gemma-4-26B-A4B-StyleTuneMoEI1-IQ1_M26.5B8.30 GiB0.13 GiB8.96 GiB0.04 GiB11±8.3%
gemma-4-26b-a4b-heretic-styletune-v2-headMoEI1-IQ1_M25.8B8.30 GiB0.13 GiB8.96 GiB0.04 GiB11±8.3%
Qianfan-OCRBF164.7B8.24 GiB0.16 GiB8.96 GiB0.04 GiB11±8.3%
Fallen-Gemma3-27B-v1IQ2_S27.4B8.18 GiB0.19 GiB8.96 GiB0.04 GiB12±8.3%
Tini-Cybersec-8B-A1BMoEQ8_08.5B8.39 GiB0.01 GiB8.95 GiB0.05 GiB32±37%
LFM2.5-8B-A1B-KO-SFTMoEQ8_08.5B8.39 GiB0.01 GiB8.95 GiB0.05 GiB32±37%
LFM2.5-8B-A1B-SOMPOA-heresyMoEQ8_08.5B8.39 GiB0.01 GiB8.95 GiB0.05 GiB32±37%
Huihui-LFM2.5-8B-A1B-abliteratedMoEQ8_08.5B8.39 GiB0.01 GiB8.95 GiB0.05 GiB32±37%
LFM2.5-8B-A1BMoEQ8_08.5B8.39 GiB0.01 GiB8.95 GiB0.05 GiB32±37%
Supertron2.1-8B-A1BMoEQ8_08.5B8.39 GiB0.01 GiB8.95 GiB0.05 GiB32±37%
LFM2.5-8B-A1B-hereticMoEQ8_08.5B8.39 GiB0.01 GiB8.95 GiB0.05 GiB32±37%
granite-20b-code-instruct-8kIQ3_S20.1B8.32 GiB0.00 GiB8.95 GiB0.05 GiB12±8.3%
granite-20b-code-base-8kI1-IQ3_S20.1B8.32 GiB0.00 GiB8.95 GiB0.05 GiB12±8.3%
Voxtral-Mini-4B-Realtime-2602KV unresolvedF164.4B8.27 GiB0.11 GiB8.95 GiB0.05 GiB11±8.3%
CycleGRPO-4BF164.8B8.23 GiB0.16 GiB8.94 GiB0.06 GiB11±8.3%
Jan-v3-4B-base-instructBF164.4B8.22 GiB0.16 GiB8.94 GiB0.06 GiB11±8.3%
Jan-code-4bBF164.4B8.22 GiB0.16 GiB8.94 GiB0.06 GiB11±8.3%
Magistral-Small-2509-VisionQ2_K_M24.0B8.09 GiB0.18 GiB8.94 GiB0.06 GiB12±8.3%
Qwen3-15B-A2B-BaseMoEQ4_K_S15.6B8.33 GiB0.05 GiB8.94 GiB0.06 GiB42±37%
Llama3.2-24B-A3B-II-Dark-Champion-INSTRUCT-Heretic-Abliterated-UncensoredMoEI1-Q3_K_M18.0B8.25 GiB0.12 GiB8.93 GiB0.07 GiB30±37%
Mistral-7B-v0.1KV unresolvedQ6_K7.2B8.20 GiB0.14 GiB8.93 GiB0.07 GiB12±8.3%
Phi-3-medium-4k-instructQ4_K_L14.0B8.09 GiB0.22 GiB8.92 GiB0.08 GiB12±8.3%
Qwen2.5-Coder-7B-InstructQ4_07.6B8.25 GiB0.06 GiB8.92 GiB0.08 GiB12±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?
1687 of 2118 indexed open-weight models fit a Apple M4 at 4,096 context with q4_0 KV cache, the largest being SOLAR-10.7B-Instruct-v1.0-uncensored at Q6_K. 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.