Apple · apple

Apple M2

Apple M2 has 24 GB of unified memory at 102 GB/s — about 16.74 GiB usable after driver and compositor overhead. 1948 of 2118 indexed models fit at 4K context with q8_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
LPDDR5-6400
Bandwidth
102 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 1673vision language 171video 16image 2audio asr 39audio tts 21embedding 26

What fits at 4K context

largest quantization that fits, per model · 1948 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
Trinity-MiniMoEQ5_K_M26.1B17.36 GiB0.10 GiB18.00 GiB0.00 GiB20±37%
Gemma-4-31B-Isometry-RPIQ4_XS32.7B16.40 GiB0.95 GiB17.99 GiB0.01 GiB5±8.3%
Gemma-4-Dark-Gemistry-31BIQ4_XS32.7B16.40 GiB0.95 GiB17.99 GiB0.01 GiB5±8.3%
Prosopon-31BIQ4_XS32.7B16.40 GiB0.95 GiB17.99 GiB0.01 GiB5±8.3%
Giftige-Blume-31B-v1-StyleSwapIQ4_XS32.7B16.40 GiB0.95 GiB17.99 GiB0.01 GiB5±8.3%
G4-MeroMero-31B-StyleSwapIQ4_XS32.7B16.40 GiB0.95 GiB17.99 GiB0.01 GiB5±8.3%
Gemma-4-31B-StyleTune-heretic-araIQ4_XS32.7B16.40 GiB0.95 GiB17.99 GiB0.01 GiB5±8.3%
Pantheon-Reasoning-31B-1.1IQ4_XS32.7B16.40 GiB0.95 GiB17.99 GiB0.01 GiB5±8.3%
Barcenas-StyleTune-31B-FableIQ4_XS32.1B16.40 GiB0.95 GiB17.99 GiB0.01 GiB5±8.3%
AMALIA-9B-0626-DPOBF169.2B17.05 GiB0.35 GiB17.98 GiB0.02 GiB5±8.3%
Skyfall-31B-v4.2Q4_K_S31.4B16.86 GiB0.45 GiB17.98 GiB0.02 GiB5±8.3%
Apertus-70B-Instruct-2509UD-IQ1_M70.6B16.58 GiB0.66 GiB17.97 GiB0.03 GiB5±8.3%
EXAONE-4.5-33BIQ4_XS34.4B16.79 GiB0.53 GiB17.97 GiB0.03 GiB5±8.3%
Qwen3.6-35B-A3B-uncensored-heretic-Native-MTP-PreservedMoEQ3_K_L35.1B17.37 GiB0.04 GiB17.97 GiB0.03 GiB26±37%
Qwen3.5-35B-A3B-uncensored-heretic-v2-Native-MTP-PreservedMoEQ3_K_L35.1B17.37 GiB0.04 GiB17.97 GiB0.03 GiB26±37%
Wan2.1-VACE-14BQ8_017.3B17.38 GiB0.00 GiB17.97 GiB0.03 GiB5±8.3%
Huihui-Qwen3.5-35B-A3B-abliteratedMoEI1-IQ4_XS36.0B17.37 GiB0.04 GiB17.97 GiB0.03 GiB26±37%
Qwen3.5-35B-A3B-BaseMoEI1-IQ4_XS36.0B17.37 GiB0.04 GiB17.97 GiB0.03 GiB26±37%
Qwen3.5-35B-A3B-Claude-4.6-Opus-Reasoning-DistilledMoEI1-IQ4_XS36.0B17.37 GiB0.04 GiB17.97 GiB0.03 GiB26±37%
Qwen3-42B-A3B-2507-Thinking-Abliterated-uncensored-TOTAL-RECALL-v2-Medium-MASTER-CODERMoEI1-IQ3_S42.4B17.14 GiB0.28 GiB17.96 GiB0.04 GiB19±37%
grug-27bQ4_K_L27.4B17.21 GiB0.13 GiB17.95 GiB0.05 GiB5±8.3%
Carnice-V2-27bQ4_K_L27.4B17.21 GiB0.13 GiB17.95 GiB0.05 GiB5±8.3%
Fara1.5-27BQ4_K_L27.4B17.21 GiB0.13 GiB17.95 GiB0.05 GiB5±8.3%
InternVL3_5-30B-A3BQ4_K_M30.8B17.35 GiB0.00 GiB17.95 GiB0.05 GiB5±8.3%
GLM-4.7-Flash-hereticMoEQ4_K_M29.9B17.27 GiB0.11 GiB17.94 GiB0.06 GiB20±37%
Qwen3.6-40B-Claude-4.6-Opus-Deckard-Heretic-Uncensored-ThinkingIQ3_M39.5B17.12 GiB0.20 GiB17.93 GiB0.07 GiB5±8.3%
Phi-3.5-MoE-instructMoEKV unresolvedIQ3_M41.9B17.11 GiB0.27 GiB17.93 GiB0.07 GiB14±37%
Gemma-3-27B-MeditronFOI1-Q4_128.8B16.81 GiB0.49 GiB17.92 GiB0.08 GiB5±8.3%
IQuest-Coder-V1-40B-InstructI1-IQ3_M39.8B16.61 GiB0.66 GiB17.92 GiB0.08 GiB5±8.3%
G4-MeroMero-26B-A4B-it-uncensored-hereticMoEQ5_K_S25.8B17.14 GiB0.24 GiB17.92 GiB0.08 GiB5±8.3%
deepseek-coder-33b-instructIQ4_XS33.3B16.77 GiB0.51 GiB17.91 GiB0.09 GiB5±8.3%
gemma-4-E2B-it-Uncensored-MAXF325.1B17.33 GiB0.03 GiB17.90 GiB0.10 GiB5±8.3%
Aurora-Code-1MoEI1-Q4_K_M34.7B17.28 GiB0.04 GiB17.88 GiB0.12 GiB27±37%
OmniAtlas-Qwen3-30B-A3BI1-Q4_K_M31.7B17.28 GiB0.00 GiB17.88 GiB0.12 GiB5±8.3%
Qwen3-Omni-30B-A3B-InstructQ4_K_M35.3B17.28 GiB0.00 GiB17.88 GiB0.12 GiB5±8.3%
Qwen3-Omni-30B-A3B-CaptionerI1-Q4_K_M31.7B17.28 GiB0.00 GiB17.88 GiB0.12 GiB5±8.3%
Qwen3-Omni-30B-A3B-ThinkingQ4_K_M31.7B17.28 GiB0.00 GiB17.88 GiB0.12 GiB5±8.3%
Qwen3.6-35B-A3B-Fable-5-DistillMoEI1-Q3_K_L36.0B17.28 GiB0.04 GiB17.87 GiB0.13 GiB27±37%
Qwable-v2MoEI1-Q3_K_L36.0B17.28 GiB0.04 GiB17.87 GiB0.13 GiB27±37%
Salience-1.5-ProMoEI1-Q3_K_L36.0B17.28 GiB0.04 GiB17.87 GiB0.13 GiB27±37%
Qwen3.6-35B-A3B-YOYO-V2MoEI1-Q3_K_L36.0B17.28 GiB0.04 GiB17.87 GiB0.13 GiB27±37%
Ornith-1.0-35B-FP8-BLOCK-MTPMoEI1-Q3_K_L35.5B17.28 GiB0.04 GiB17.87 GiB0.13 GiB27±37%
fable-coder-35B-A3BMoEI1-Q3_K_L36.0B17.28 GiB0.04 GiB17.87 GiB0.13 GiB27±37%
Qwen3.6-35B-A3B-AntiLoopMoEI1-Q3_K_L36.0B17.28 GiB0.04 GiB17.87 GiB0.13 GiB27±37%
PINQWEN-3.6-35B-CLEAN-BF16MoEI1-Q3_K_L36.0B17.28 GiB0.04 GiB17.87 GiB0.13 GiB27±37%
UniMath-35B-A3BMoEI1-Q3_K_L36.0B17.28 GiB0.04 GiB17.87 GiB0.13 GiB27±37%
Ornith-1.0-35B-Heretic-MTPMoEI1-Q3_K_L—17.28 GiB0.04 GiB17.87 GiB0.13 GiB27±37%
Fawen-1.0-35BMoEI1-Q3_K_L36.0B17.28 GiB0.04 GiB17.87 GiB0.13 GiB27±37%
CyberStrike-OffSec-35BMoEQ3_K_L35.1B17.28 GiB0.04 GiB17.87 GiB0.13 GiB27±37%
WizardCoder-Python-34B-V1.0I1-IQ4_XS33.7B16.83 GiB0.40 GiB17.87 GiB0.13 GiB5±8.3%
Phind-CodeLlama-34B-Python-v1I1-IQ4_XS33.7B16.83 GiB0.40 GiB17.87 GiB0.13 GiB5±8.3%
Phind-CodeLlama-34B-v2I1-IQ4_XS33.7B16.83 GiB0.40 GiB17.87 GiB0.13 GiB5±8.3%
Qwen3.6-35B-A3BMoEQ3_K_L36.0B17.28 GiB0.04 GiB17.87 GiB0.13 GiB27±37%
Gemma-4-Novelist-Eclipse-31BI1-IQ4_XS32.7B16.28 GiB0.95 GiB17.87 GiB0.13 GiB5±8.3%
Gemma-4-31B-StyleTuneI1-IQ4_XS32.7B16.28 GiB0.95 GiB17.87 GiB0.13 GiB5±8.3%
GLM-Z1-Rumination-32B-0414IQ4_XS33.1B16.72 GiB0.51 GiB17.86 GiB0.14 GiB5±8.3%
GRM-2.6-Plus-0628Q4_K_M27.8B17.12 GiB0.13 GiB17.86 GiB0.14 GiB5±8.3%
Qwen3.6-35B-A3BMoEUD-IQ4_NL36.0B17.26 GiB0.04 GiB17.86 GiB0.14 GiB27±37%
Marco-Mini-InstructMoEQ8_017.3B17.10 GiB0.23 GiB17.85 GiB0.15 GiB24±37%
dolphin-2.6-mixtral-8x7bMoEI1-IQ3_XXS46.7B16.99 GiB0.27 GiB17.84 GiB0.16 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 run?
1948 of 2118 indexed open-weight models fit a Apple M2 at 4,096 context with q8_0 KV cache, the largest being Trinity-Mini at Q5_K_M. That covers text, vision-language, image, video and speech models.
How much usable memory does a Apple M2 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 M2 fast for local AI?
Its memory bandwidth is 102 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.