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. 1951 of 2118 indexed models fit at 8K 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
text 1674vision language 173video 16image 2audio asr 39audio tts 21embedding 26

What fits at 8K context

largest quantization that fits, per model · 1951 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
EXAONE-4.0-32BQ4_032.0B16.96 GiB0.38 GiB17.99 GiB0.01 GiB6±8.3%
Gemma-4-Novelist-Eclipse-31BIQ4_XS32.7B16.67 GiB0.68 GiB17.99 GiB0.01 GiB6±8.3%
Gemma-4-31B-StyleTuneIQ4_XS32.7B16.67 GiB0.68 GiB17.99 GiB0.01 GiB6±8.3%
Qwen3-42B-A3B-2507-Thinking-Abliterated-uncensored-TOTAL-RECALL-v2-Medium-MASTER-CODERMoEI1-IQ3_S42.4B17.14 GiB0.29 GiB17.98 GiB0.02 GiB22±37%
Trinity-MiniMoEQ5_K_M26.1B17.36 GiB0.07 GiB17.97 GiB0.03 GiB24±37%
Qwen3.6-35B-A3B-uncensored-heretic-Native-MTP-PreservedMoEQ3_K_L35.1B17.37 GiB0.04 GiB17.97 GiB0.03 GiB30±37%
Qwen3.5-35B-A3B-uncensored-heretic-v2-Native-MTP-PreservedMoEQ3_K_L35.1B17.37 GiB0.04 GiB17.97 GiB0.03 GiB30±37%
Huihui-Qwen3.5-35B-A3B-abliteratedMoEI1-IQ4_XS36.0B17.37 GiB0.04 GiB17.97 GiB0.03 GiB30±37%
Qwen3.5-35B-A3B-BaseMoEI1-IQ4_XS36.0B17.37 GiB0.04 GiB17.97 GiB0.03 GiB30±37%
Qwen3.5-35B-A3B-Claude-4.6-Opus-Reasoning-DistilledMoEI1-IQ4_XS36.0B17.37 GiB0.04 GiB17.97 GiB0.03 GiB30±37%
Wan2.1-VACE-14BQ8_017.3B17.38 GiB0.00 GiB17.97 GiB0.03 GiB6±8.3%
grug-27bQ4_K_L27.4B17.21 GiB0.14 GiB17.96 GiB0.04 GiB6±8.3%
Carnice-V2-27bQ4_K_L27.4B17.21 GiB0.14 GiB17.96 GiB0.04 GiB6±8.3%
Fara1.5-27BQ4_K_L27.4B17.21 GiB0.14 GiB17.96 GiB0.04 GiB6±8.3%
IQuest-Coder-V1-40B-InstructI1-IQ3_M39.8B16.61 GiB0.70 GiB17.96 GiB0.04 GiB6±8.3%
InternVL3_5-30B-A3BQ4_K_M30.8B17.35 GiB0.00 GiB17.95 GiB0.05 GiB6±8.3%
Qwen3.6-40B-Claude-4.6-Opus-Deckard-Heretic-Uncensored-ThinkingIQ3_M39.5B17.12 GiB0.21 GiB17.94 GiB0.06 GiB6±8.3%
Phi-3.5-MoE-instructMoEKV unresolvedIQ3_M41.9B17.11 GiB0.28 GiB17.94 GiB0.06 GiB16±37%
deepseek-coder-33b-instructIQ4_XS33.3B16.77 GiB0.54 GiB17.94 GiB0.06 GiB6±8.3%
GLM-4.7-Flash-hereticMoEQ4_K_M29.9B17.27 GiB0.12 GiB17.94 GiB0.06 GiB23±37%
Skyfall-31B-v4.2IQ4_NL31.4B16.79 GiB0.47 GiB17.94 GiB0.06 GiB6±8.3%
Goetia-26B-A4B-v1.4MoEI1-Q5_K_S26.0B17.22 GiB0.17 GiB17.93 GiB0.07 GiB6±8.3%
G4-Moonlight-Dusk-26B-A4B-hereticMoEI1-Q5_K_S26.5B17.22 GiB0.17 GiB17.93 GiB0.07 GiB6±8.3%
Pantheon-Reasoning-26B-A4B-1.1-hereticMoEI1-Q5_K_S26.5B17.22 GiB0.17 GiB17.93 GiB0.07 GiB6±8.3%
G4-Moonlight-Dusk-26B-A4BMoEI1-Q5_K_S26.5B17.22 GiB0.17 GiB17.93 GiB0.07 GiB6±8.3%
Chimera-X-26B-A4BMoEI1-Q5_K_S26.5B17.22 GiB0.17 GiB17.93 GiB0.07 GiB6±8.3%
Pantheon-Reasoning-26B-A4B-1.1MoEI1-Q5_K_S26.5B17.22 GiB0.17 GiB17.93 GiB0.07 GiB6±8.3%
Gemma-4-26B-A4B-StyleTune-V2MoEI1-Q5_K_S26.5B17.22 GiB0.17 GiB17.93 GiB0.07 GiB6±8.3%
Gemma-4-26B-A4B-StyleTuneMoEI1-Q5_K_S26.5B17.22 GiB0.17 GiB17.93 GiB0.07 GiB6±8.3%
gemma-4-26b-a4b-heretic-styletune-v2-headMoEI1-Q5_K_S25.8B17.22 GiB0.17 GiB17.93 GiB0.07 GiB6±8.3%
Wizard-Vicuna-30B-UncensoredI1-IQ3_M32.5B13.86 GiB3.43 GiB17.91 GiB0.09 GiB6±8.3%
archangel_sft-kto_llama30bI1-IQ3_M32.5B13.86 GiB3.43 GiB17.91 GiB0.09 GiB6±8.3%
WizardCoder-Python-34B-V1.0I1-IQ4_XS33.7B16.83 GiB0.42 GiB17.90 GiB0.10 GiB6±8.3%
Phind-CodeLlama-34B-Python-v1I1-IQ4_XS33.7B16.83 GiB0.42 GiB17.90 GiB0.10 GiB6±8.3%
Phind-CodeLlama-34B-v2I1-IQ4_XS33.7B16.83 GiB0.42 GiB17.90 GiB0.10 GiB6±8.3%
gemma-4-E2B-it-Uncensored-MAXF325.1B17.33 GiB0.02 GiB17.89 GiB0.11 GiB6±8.3%
GLM-Z1-Rumination-32B-0414IQ4_XS33.1B16.72 GiB0.54 GiB17.89 GiB0.11 GiB6±8.3%
Aurora-Code-1MoEI1-Q4_K_M34.7B17.28 GiB0.04 GiB17.88 GiB0.12 GiB30±37%
OmniAtlas-Qwen3-30B-A3BI1-Q4_K_M31.7B17.28 GiB0.00 GiB17.88 GiB0.12 GiB6±8.3%
Qwen3-Omni-30B-A3B-InstructQ4_K_M35.3B17.28 GiB0.00 GiB17.88 GiB0.12 GiB6±8.3%
Qwen3-Omni-30B-A3B-CaptionerI1-Q4_K_M31.7B17.28 GiB0.00 GiB17.88 GiB0.12 GiB6±8.3%
Qwen3-Omni-30B-A3B-ThinkingQ4_K_M31.7B17.28 GiB0.00 GiB17.88 GiB0.12 GiB6±8.3%
Qwen3.6-35B-A3B-Fable-5-DistillMoEI1-Q3_K_L36.0B17.28 GiB0.04 GiB17.88 GiB0.12 GiB30±37%
Qwable-v2MoEI1-Q3_K_L36.0B17.28 GiB0.04 GiB17.88 GiB0.12 GiB30±37%
Salience-1.5-ProMoEI1-Q3_K_L36.0B17.28 GiB0.04 GiB17.88 GiB0.12 GiB30±37%
Qwen3.6-35B-A3B-YOYO-V2MoEI1-Q3_K_L36.0B17.28 GiB0.04 GiB17.88 GiB0.12 GiB30±37%
Ornith-1.0-35B-FP8-BLOCK-MTPMoEI1-Q3_K_L35.5B17.28 GiB0.04 GiB17.88 GiB0.12 GiB30±37%
fable-coder-35B-A3BMoEI1-Q3_K_L36.0B17.28 GiB0.04 GiB17.88 GiB0.12 GiB30±37%
Qwen3.6-35B-A3B-AntiLoopMoEI1-Q3_K_L36.0B17.28 GiB0.04 GiB17.88 GiB0.12 GiB30±37%
PINQWEN-3.6-35B-CLEAN-BF16MoEI1-Q3_K_L36.0B17.28 GiB0.04 GiB17.88 GiB0.12 GiB30±37%
UniMath-35B-A3BMoEI1-Q3_K_L36.0B17.28 GiB0.04 GiB17.88 GiB0.12 GiB30±37%
Ornith-1.0-35B-Heretic-MTPMoEI1-Q3_K_L—17.28 GiB0.04 GiB17.88 GiB0.12 GiB30±37%
Fawen-1.0-35BMoEI1-Q3_K_L36.0B17.28 GiB0.04 GiB17.88 GiB0.12 GiB30±37%
CyberStrike-OffSec-35BMoEQ3_K_L35.1B17.28 GiB0.04 GiB17.88 GiB0.12 GiB30±37%
Qwen3.6-35B-A3BMoEQ3_K_L36.0B17.28 GiB0.04 GiB17.88 GiB0.12 GiB30±37%
GRM-2.6-Plus-0628Q4_K_M27.8B17.12 GiB0.14 GiB17.87 GiB0.13 GiB6±8.3%
Marco-Mini-InstructMoEQ8_017.3B17.10 GiB0.25 GiB17.87 GiB0.13 GiB28±37%
Qwen3.6-35B-A3BMoEUD-IQ4_NL36.0B17.26 GiB0.04 GiB17.86 GiB0.14 GiB30±37%
dolphin-2.6-mixtral-8x7bMoEI1-IQ3_XXS46.7B16.99 GiB0.28 GiB17.86 GiB0.14 GiB10±37%
xLAM-8x7b-rMoEIQ3_XXS46.7B16.99 GiB0.28 GiB17.86 GiB0.14 GiB10±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 M4 run?
1951 of 2118 indexed open-weight models fit a Apple M4 at 8,192 context with q4_0 KV cache, the largest being EXAONE-4.0-32B at Q4_0. 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.