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. 1535 of 2118 indexed models fit at 32K 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 1321video 12vision language 114image 2audio asr 39audio tts 21embedding 26

What fits at 32K context

largest quantization that fits, per model · 1535 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
Le-Chaton-Slim-23BMoEI1-Q2_K_S23.3B7.52 GiB0.91 GiB9.00 GiB0.00 GiB19±37%
Wan2.1-T2V-14BQ4_014.3B8.41 GiB0.00 GiB9.00 GiB0.00 GiB11±8.3%
GLM-4.7-Flash-REAP-23B-A3BMoEUD-IQ2_M23.0B7.97 GiB0.46 GiB8.99 GiB0.01 GiB31±37%
Qwen3-VL-30B-A3B-InstructMoEUD-TQ1_031.1B7.60 GiB0.84 GiB8.99 GiB0.01 GiB28±37%
GigaChat3-10B-A1.8B-baseMoEQ6_K11.5B8.18 GiB0.26 GiB8.99 GiB0.01 GiB37±37%
Falcon3-7B-InstructQ8_07.5B7.38 GiB0.98 GiB8.98 GiB0.02 GiB12±8.3%
Gemma-The-Writer-N-Restless-Quill-10B-UncensoredI1-Q5_K_S10.0B6.55 GiB1.84 GiB8.98 GiB0.02 GiB11±8.3%
gemma-3n-E2B-itF165.4B8.31 GiB0.12 GiB8.98 GiB0.02 GiB11±8.3%
InternVL3_5-14BQ4_K_M15.1B8.38 GiB0.00 GiB8.98 GiB0.02 GiB11±8.3%
GLM-4.7-Flash-hereticMoEIQ2_XXS29.9B7.95 GiB0.46 GiB8.98 GiB0.02 GiB33±37%
Phi-3-mini-4k-instructKV unresolvedIQ3_XXS3.8B5.05 GiB3.38 GiB8.98 GiB0.02 GiB11±8.3%
Qwen3-VL-30B-A3B-ThinkingMoEUD-TQ1_031.1B7.59 GiB0.84 GiB8.98 GiB0.02 GiB28±37%
Qwen3-30B-A3B-Thinking-2507MoEUD-TQ1_030.5B7.59 GiB0.84 GiB8.98 GiB0.02 GiB28±37%
Kimi-VL-A3B-InstructMoEI1-IQ4_XS16.4B8.15 GiB0.27 GiB8.98 GiB0.02 GiB32±37%
Moonlight-16B-A3B-InstructMoEIQ4_XS16.0B8.15 GiB0.27 GiB8.98 GiB0.02 GiB32±37%
Ministral-3-14B-Instruct-2512-BF16-abliteratedIQ4_XS13.9B6.96 GiB1.41 GiB8.97 GiB0.03 GiB12±8.3%
Ministral-3-14B-abliteratedIQ4_XS13.9B6.96 GiB1.41 GiB8.97 GiB0.03 GiB12±8.3%
Ministral-3-14B-Reasoning-2512-UncensoredIQ4_XS13.9B6.96 GiB1.41 GiB8.97 GiB0.03 GiB12±8.3%
Qwen3-30B-A3BMoEIQ2_XXS30.5B7.59 GiB0.84 GiB8.97 GiB0.03 GiB28±37%
Pantheon-Proto-RP-1.8-30B-A3BMoEIQ2_XXS30.5B7.59 GiB0.84 GiB8.97 GiB0.03 GiB28±37%
HunyuanVideo-1.5Q8_08.3B8.38 GiB0.00 GiB8.97 GiB0.03 GiB12±8.3%
Forsaken-Void-12BI1-Q4_K_M12.2B6.96 GiB1.41 GiB8.97 GiB0.03 GiB12±8.3%
Silver-Siren-ST-12BI1-Q4_K_M12.2B6.96 GiB1.41 GiB8.97 GiB0.03 GiB12±8.3%
Tess-3-Mistral-Nemo-12BI1-Q4_K_M12.2B6.96 GiB1.41 GiB8.97 GiB0.03 GiB12±8.3%
KrakenSakura-Maelstrom-12B-v1Q4_K_M12.2B6.96 GiB1.41 GiB8.97 GiB0.03 GiB12±8.3%
MN-12B-Runeweaver-RP-RUI1-Q4_K_M12.2B6.96 GiB1.41 GiB8.97 GiB0.03 GiB12±8.3%
Impish_Bloodmoon_12BI1-Q4_K_M12.2B6.96 GiB1.41 GiB8.97 GiB0.03 GiB12±8.3%
Wayfarer-2-12BI1-Q4_K_M12.2B6.96 GiB1.41 GiB8.97 GiB0.03 GiB12±8.3%
Wayfarer-12BI1-Q4_K_M12.2B6.96 GiB1.41 GiB8.97 GiB0.03 GiB12±8.3%
Muse-12BI1-Q4_K_M12.2B6.96 GiB1.41 GiB8.97 GiB0.03 GiB12±8.3%
Vikhr-Nemo-12B-Instruct-R-21-09-24Q4_K_M12.2B6.96 GiB1.41 GiB8.97 GiB0.03 GiB12±8.3%
Mistral-Nemo-Base-2407Q4_K_M12.2B6.96 GiB1.41 GiB8.97 GiB0.03 GiB12±8.3%
writing-roleplay-20k-context-nemo-12b-v1.0Q4_K_M12.2B6.96 GiB1.41 GiB8.97 GiB0.03 GiB12±8.3%
Dans-PersonalityEngine-V1.3.0-12bI1-Q4_K_M12.2B6.96 GiB1.41 GiB8.97 GiB0.03 GiB12±8.3%
mini-magnum-12b-v1.1Q4_K_M12.2B6.96 GiB1.41 GiB8.97 GiB0.03 GiB12±8.3%
Lumimaid-v0.2-12BQ4_K_M12.2B6.96 GiB1.41 GiB8.97 GiB0.03 GiB12±8.3%
MN-Violet-Lotus-12BI1-Q4_K_M12.2B6.96 GiB1.41 GiB8.97 GiB0.03 GiB12±8.3%
Rocinante-X-12B-v1-Heretic-UncensoredI1-Q4_K_M12.2B6.96 GiB1.41 GiB8.97 GiB0.03 GiB12±8.3%
Mistral-Heretica-12BI1-Q4_K_M12.2B6.96 GiB1.41 GiB8.97 GiB0.03 GiB12±8.3%
Violet_Twilight-v0.2Q4_K_M12.2B6.96 GiB1.41 GiB8.97 GiB0.03 GiB12±8.3%
Lumimaid-Magnum-v4-12BQ4_K_M12.2B6.96 GiB1.41 GiB8.97 GiB0.03 GiB12±8.3%
arcee-fusion-lumaid-12BI1-Q4_K_M12.2B6.96 GiB1.41 GiB8.97 GiB0.03 GiB12±8.3%
Mistral-NeMo-12B-AbliteratedI1-Q4_K_M12.2B6.96 GiB1.41 GiB8.97 GiB0.03 GiB12±8.3%
Captain-Eris_Violet-V0.420-12BI1-Q4_K_M12.2B6.96 GiB1.41 GiB8.97 GiB0.03 GiB12±8.3%
Rocinante-X-12B-v1-absolute-heresyI1-Q4_K_M12.2B6.96 GiB1.41 GiB8.97 GiB0.03 GiB12±8.3%
Rocinante-X-12B-v1I1-Q4_K_M12.2B6.96 GiB1.41 GiB8.97 GiB0.03 GiB12±8.3%
Mistral-Nemo-Gutenberg-Doppel-12BQ4_K_M12.2B6.96 GiB1.41 GiB8.97 GiB0.03 GiB12±8.3%
Mistral-Nemo-Instruct-2407Q4_K_M12.2B6.96 GiB1.41 GiB8.97 GiB0.03 GiB12±8.3%
magnum-v4-12bQ4_K_M12.2B6.96 GiB1.41 GiB8.97 GiB0.03 GiB12±8.3%
MN-12b-RP-InkQ4_K_M12.2B6.96 GiB1.41 GiB8.97 GiB0.03 GiB12±8.3%
Mistral-Nemo-12B-ArliAI-RPMax-v1.1Q4_K_M12.2B6.96 GiB1.41 GiB8.97 GiB0.03 GiB12±8.3%
Mistral-Nemo-2407-12B-Thinking-Claude-Gemini-GPT5.2-Uncensored-HERETICI1-Q4_K_M12.2B6.96 GiB1.41 GiB8.97 GiB0.03 GiB12±8.3%
Dans-SakuraKaze-V1.0.0-12bI1-Q4_K_M12.2B6.96 GiB1.41 GiB8.97 GiB0.03 GiB12±8.3%
Mistral-Nemo-Inst-2407-12B-Thinking-Uncensored-HERETIC-HI-Claude-OpusI1-Q4_K_M12.2B6.96 GiB1.41 GiB8.97 GiB0.03 GiB12±8.3%
Mistral-Nemo-Instruct-2407-12B-Thinking-M-Claude-Opus-High-ReasoningI1-Q4_K_M12.2B6.96 GiB1.41 GiB8.97 GiB0.03 GiB12±8.3%
MN-12B-Mag-Mell-R1Q4_K_M12.2B6.96 GiB1.41 GiB8.97 GiB0.03 GiB12±8.3%
Mordant-12B-ThinkI1-Q4_K_M12.2B6.96 GiB1.41 GiB8.97 GiB0.03 GiB12±8.3%
Riverfish-Rocinante-12B-SFT-DPOI1-Q4_K_M12.2B6.96 GiB1.41 GiB8.97 GiB0.03 GiB12±8.3%
MN-Violet-Lotus-12B-HereticI1-Q4_K_M12.2B6.96 GiB1.41 GiB8.97 GiB0.03 GiB12±8.3%
Himeyuri-Magnum-12B-HereticMergeI1-Q4_K_M12.2B6.96 GiB1.41 GiB8.97 GiB0.03 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?
1535 of 2118 indexed open-weight models fit a Apple M4 at 32,768 context with q4_0 KV cache, the largest being Le-Chaton-Slim-23B at I1-Q2_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.