Apple · apple

Apple M1

Apple M1 has 16 GB of unified memory at 68 GB/s — about 11.16 GiB usable after driver and compositor overhead. 1820 of 2118 indexed models fit at 8K context with q4_0 KV. Note only 12 GB of its 16 GB is allocatable to the GPU.

Spec sheet· bandwidth, theoreticalFrom the file· fit from summed bytesPredicted· speed
Memory
16 GB
LPDDR4X-4266
Bandwidth
68 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 1566vision language 151video 15audio asr 39audio tts 21image 2embedding 26

What fits at 8K context

largest quantization that fits, per model · 1820 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
Qwen3.5-40B-RoughHouse-Claude-4.6-Opus-Polar-Deckard-Uncensored-Heretic-ThinkingI1-IQ2_XXS39.5B11.17 GiB0.21 GiB12.00 GiB0.00 GiB5±8.3%
Qwen3.5-27B-Engineer-Deckard-GeminiI1-Q3_K_S27.7B11.24 GiB0.14 GiB12.00 GiB0.00 GiB5±8.3%
Qwen3.5-27B-HERETIC-Polaris-Advanced-Thinking-Alpha-uncensoredI1-Q3_K_S27.4B11.24 GiB0.14 GiB12.00 GiB0.00 GiB5±8.3%
Qwen3.5-27B-Deckard-PKD-Heretic-Uncensored-ThinkingI1-Q3_K_S27.4B11.24 GiB0.14 GiB12.00 GiB0.00 GiB5±8.3%
Qwen3.6-27B-Heretic2-ThinkingI1-Q3_K_S27.4B11.24 GiB0.14 GiB12.00 GiB0.00 GiB5±8.3%
Qwen3.6-27B-Uncensored-AggressiveI1-Q3_K_S27.4B11.24 GiB0.14 GiB12.00 GiB0.00 GiB5±8.3%
Qwen-3.5-Opus-GLM-27BI1-Q3_K_S26.9B11.24 GiB0.14 GiB12.00 GiB0.00 GiB5±8.3%
Qwen3.6-27B-abliteratedI1-Q3_K_S27.4B11.24 GiB0.14 GiB12.00 GiB0.00 GiB5±8.3%
KoQweopus-3.5-27B-experimentalI1-Q3_K_S27.8B11.24 GiB0.14 GiB12.00 GiB0.00 GiB5±8.3%
Webcoda-AI-27BI1-Q3_K_S27.4B11.24 GiB0.14 GiB12.00 GiB0.00 GiB5±8.3%
Qwen3.5-27B-imabari-v2I1-Q3_K_S27.8B11.24 GiB0.14 GiB12.00 GiB0.00 GiB5±8.3%
Qwen3.6-27B-AEON-Ultimate-Uncensored-BF16Q3_K_S27.4B11.24 GiB0.14 GiB12.00 GiB0.00 GiB5±8.3%
Huihui-Qwen3.5-27B-abliteratedI1-Q3_K_S27.8B11.24 GiB0.14 GiB12.00 GiB0.00 GiB5±8.3%
Qwen3.5-27B-uncensored-heretic-v1I1-Q3_K_S27.4B11.24 GiB0.14 GiB12.00 GiB0.00 GiB5±8.3%
Qwen3.5-27B-Unredacted-MAXI1-Q3_K_S27.4B11.24 GiB0.14 GiB12.00 GiB0.00 GiB5±8.3%
Qwen3.5-27B-hereticI1-Q3_K_S27.4B11.24 GiB0.14 GiB12.00 GiB0.00 GiB5±8.3%
Carnice-V2-27bI1-Q3_K_S27.4B11.24 GiB0.14 GiB12.00 GiB0.00 GiB5±8.3%
Qwen3.5-Queen-27BI1-Q3_K_S27.4B11.24 GiB0.14 GiB12.00 GiB0.00 GiB5±8.3%
GRaPE-2-ProI1-Q3_K_S27.8B11.24 GiB0.14 GiB12.00 GiB0.00 GiB5±8.3%
Huihui-Qwen3.6-27B-abliteratedQ3_K_S27.8B11.24 GiB0.14 GiB12.00 GiB0.00 GiB5±8.3%
Qwen3.5-27B-abliteratedQ3_K_S26.9B11.24 GiB0.14 GiB12.00 GiB0.00 GiB5±8.3%
Qwen3.5-27B-DerestrictedI1-Q3_K_S27.8B11.24 GiB0.14 GiB12.00 GiB0.00 GiB5±8.3%
ThinkingCap-Qwen3.6-27B-hereticQ3_K_S27.4B11.24 GiB0.14 GiB12.00 GiB0.00 GiB5±8.3%
Qwen3.6-27B-Claude-Opus-Reasoning-Distill-v2-hereticQ3_K_S27.4B11.24 GiB0.14 GiB12.00 GiB0.00 GiB5±8.3%
Darwin-28B-REASONI1-Q3_K_S26.9B11.24 GiB0.14 GiB12.00 GiB0.00 GiB5±8.3%
Huihui-Qwen3.5-27B-Claude-4.6-Opus-abliteratedI1-Q3_K_S27.8B11.24 GiB0.14 GiB12.00 GiB0.00 GiB5±8.3%
Qwen3.5-27B-Claude-4.6-Opus-Reasoning-DistilledI1-Q3_K_S27.8B11.24 GiB0.14 GiB12.00 GiB0.00 GiB5±8.3%
Qwen3.5-27B-WebNovel-Writer-zhI1-Q3_K_S26.9B11.24 GiB0.14 GiB12.00 GiB0.00 GiB5±8.3%
Qwen3.5-27B_Homebrew-v2I1-Q3_K_S27.4B11.24 GiB0.14 GiB12.00 GiB0.00 GiB5±8.3%
Wan2.1-VACE-14BQ5_K_S17.3B11.41 GiB0.00 GiB12.00 GiB0.00 GiB5±8.3%
Laguna-XS-2.1MoEQ2_K33.4B11.32 GiB0.12 GiB11.99 GiB0.01 GiB25±37%
command-r-35b-writer-v2I1-IQ1_M35.0B8.52 GiB2.81 GiB11.99 GiB0.01 GiB5±8.3%
MN-GRAND-23.5B-Gutenberg-UNCENSORED-V2-GLM4.7-ThinkingI1-Q3_K_M23.4B10.67 GiB0.71 GiB11.98 GiB0.02 GiB5±8.3%
Pantheon-Reasoning-27BQ2_K27.8B11.22 GiB0.14 GiB11.98 GiB0.02 GiB5±8.3%
Qwen3.5-27BQ2_K27.8B11.22 GiB0.14 GiB11.98 GiB0.02 GiB5±8.3%
ALIA-40b-fc-2606I1-IQ2_XXS40.4B10.89 GiB0.42 GiB11.97 GiB0.03 GiB5±8.3%
ALIA-40b-instruct-2606I1-IQ2_XXS40.4B10.89 GiB0.42 GiB11.97 GiB0.03 GiB5±8.3%
InternVL3_5-30B-A3BIQ3_XXS30.8B11.38 GiB0.00 GiB11.97 GiB0.03 GiB5±8.3%
Qwen3.5-21B-Claude-4.6-Opus-Deckard-Heretic-Uncensored-ThinkingI1-Q4_021.3B11.25 GiB0.11 GiB11.97 GiB0.03 GiB5±8.3%
Qwen3.6-21B-IQ-Ultra-Heretic-Uncensored-ThinkingI1-Q4_021.3B11.25 GiB0.11 GiB11.97 GiB0.03 GiB5±8.3%
Gemma-4-31B-Isometry-RPI1-Q2_K_S32.7B10.65 GiB0.68 GiB11.97 GiB0.03 GiB5±8.3%
Gemma-4-Dark-Gemistry-31BI1-Q2_K_S32.7B10.65 GiB0.68 GiB11.97 GiB0.03 GiB5±8.3%
Prosopon-31BI1-Q2_K_S32.7B10.65 GiB0.68 GiB11.97 GiB0.03 GiB5±8.3%
Gemma-4-Novelist-Eclipse-31BI1-Q2_K_S32.7B10.65 GiB0.68 GiB11.97 GiB0.03 GiB5±8.3%
Giftige-Blume-31B-v1-StyleSwapI1-Q2_K_S32.7B10.65 GiB0.68 GiB11.97 GiB0.03 GiB5±8.3%
G4-MeroMero-31B-StyleSwapI1-Q2_K_S32.7B10.65 GiB0.68 GiB11.97 GiB0.03 GiB5±8.3%
Gemma-4-31B-StyleTune-heretic-araI1-Q2_K_S32.7B10.65 GiB0.68 GiB11.97 GiB0.03 GiB5±8.3%
Pantheon-Reasoning-31B-1.1I1-Q2_K_S32.7B10.65 GiB0.68 GiB11.97 GiB0.03 GiB5±8.3%
Gemma-4-31B-StyleTuneI1-Q2_K_S32.7B10.65 GiB0.68 GiB11.97 GiB0.03 GiB5±8.3%
Barcenas-StyleTune-31B-FableI1-Q2_K_S32.1B10.65 GiB0.68 GiB11.97 GiB0.03 GiB5±8.3%
Qwen3-VL-32B-InstructUD-IQ2_M33.4B10.76 GiB0.56 GiB11.96 GiB0.04 GiB5±8.3%
Qwen3-VL-32B-ThinkingUD-IQ2_M33.4B10.76 GiB0.56 GiB11.96 GiB0.04 GiB5±8.3%
Qwen3-32BUD-IQ2_M32.8B10.76 GiB0.56 GiB11.96 GiB0.04 GiB5±8.3%
OpenAI-gpt-oss-20B-Claude-4.5-Opus-Heretic-UncensoredMoEI1-IQ3_M20.9B11.36 GiB0.06 GiB11.96 GiB0.04 GiB15±37%
gpt-oss-20b-uncensoredMoEI1-IQ3_M20.9B11.36 GiB0.06 GiB11.96 GiB0.04 GiB15±37%
gpt-oss-safeguard-20bMoEI1-IQ3_M21.5B11.36 GiB0.06 GiB11.96 GiB0.04 GiB15±37%
Huihui-gpt-oss-20b-BF16-abliterated-v2MoEI1-IQ3_M20.9B11.36 GiB0.06 GiB11.96 GiB0.04 GiB15±37%
metatune-gpt20b-R1.09MoEI1-IQ3_M21.5B11.36 GiB0.06 GiB11.96 GiB0.04 GiB15±37%
TildeOpen-30B-Instruct-LVI1-Q2_K30.7B10.80 GiB0.53 GiB11.96 GiB0.04 GiB5±8.3%
Qwen3.5-40B-Claude-4.6-Opus-Deckard-Heretic-Uncensored-ThinkingI1-IQ2_XS39.5B11.13 GiB0.21 GiB11.96 GiB0.04 GiB5±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 M1 run?
1820 of 2118 indexed open-weight models fit a Apple M1 at 8,192 context with q4_0 KV cache, the largest being Qwen3.5-40B-RoughHouse-Claude-4.6-Opus-Polar-Deckard-Uncensored-Heretic-Thinking at I1-IQ2_XXS. That covers text, vision-language, image, video and speech models.
How much usable memory does a Apple M1 actually have?
Its nameplate is 16 GB, but about 11.16 GiB is available to a model once driver and compositor overhead is accounted for, and only 12 GB of the pool can be allocated to the GPU at all.
Is a Apple M1 fast for local AI?
Its memory bandwidth is 68 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.