Apple · apple

Apple M5

Apple M5 has 32 GB of unified memory at 154 GB/s — about 22.32 GiB usable after driver and compositor overhead. 1969 of 2118 indexed models fit at 8K context with f16 KV. Note only 24 GB of its 32 GB is allocatable to the GPU.

Spec sheet· bandwidth, theoreticalFrom the file· fit from summed bytesPredicted· speed
Memory
32 GB
LPDDR5X-9600
Bandwidth
154 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 1689vision language 176audio tts 21image 2video 16audio asr 39embedding 26

What fits at 8K context

largest quantization that fits, per model · 1969 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
Gemma-4-Novelist-Eclipse-31BQ5_K_S32.7B20.93 GiB2.42 GiB23.98 GiB0.02 GiB5±8.3%
Gemma-4-31B-StyleTuneQ5_K_S32.7B20.93 GiB2.42 GiB23.98 GiB0.02 GiB5±8.3%
InternVL3_5-30B-A3BQ6_K30.8B23.38 GiB0.00 GiB23.98 GiB0.02 GiB5±8.3%
GLM-4.7-FlashMoEQ6_K31.2B23.00 GiB0.41 GiB23.97 GiB0.03 GiB21±37%
Nemotron-Cascade-2-30B-A3BMoEQ4_K_M31.6B23.03 GiB0.41 GiB23.97 GiB0.03 GiB23±37%
OmniAtlas-Qwen3-30B-A3BI1-Q6_K31.7B23.37 GiB0.00 GiB23.96 GiB0.04 GiB5±8.3%
Qwen3-Omni-30B-A3B-CaptionerI1-Q6_K31.7B23.37 GiB0.00 GiB23.96 GiB0.04 GiB5±8.3%
Qwen3.6-40B-Claude-4.6-Opus-Deckard-Heretic-Uncensored-ThinkingQ4_K_M39.5B22.59 GiB0.75 GiB23.95 GiB0.05 GiB5±8.3%
deepseek-coder-33b-instructQ5_033.3B21.38 GiB1.94 GiB23.95 GiB0.05 GiB5±8.3%
deepseek-coder-33b-baseQ5_033.3B21.38 GiB1.94 GiB23.95 GiB0.05 GiB5±8.3%
WhiteRabbitNeo-33B-v1Q5_033.3B21.38 GiB1.94 GiB23.95 GiB0.05 GiB6±8.3%
Nemotron-Labs-Audex-30B-A3BQ4_K_L32.0B23.35 GiB0.00 GiB23.95 GiB0.05 GiB5±8.3%
Qwen-AgentWorld-35B-A3BMoEUD-Q5_K_S34.7B23.23 GiB0.16 GiB23.94 GiB0.06 GiB28±37%
Ornith-1.0-35BMoEUD-Q5_K_S34.7B23.23 GiB0.16 GiB23.94 GiB0.06 GiB28±37%
OLMo-2-0325-32BQ5_K_M32.2B21.29 GiB2.00 GiB23.94 GiB0.06 GiB6±8.3%
Qwen3-48B-A4B-Savant-Commander-Distill-12X-Closed-Open-Heretic-UncensoredMoEI1-Q5_K_M33.6B22.25 GiB1.13 GiB23.93 GiB0.07 GiB12±37%
ALIA-40b-fc-2606I1-Q4_040.4B21.76 GiB1.50 GiB23.93 GiB0.07 GiB6±8.3%
ALIA-40b-instruct-2606I1-Q4_040.4B21.76 GiB1.50 GiB23.93 GiB0.07 GiB6±8.3%
Qwen3-Coder-Next-REAMMoEI1-IQ3_XS60.3B23.19 GiB0.19 GiB23.92 GiB0.08 GiB29±37%
Qwen3-Next-80B-A3B-ThinkingMoEUD-IQ1_M81.3B22.61 GiB0.75 GiB23.90 GiB0.10 GiB25±37%
Qwen3.5-122B-A10B-hereticMoEI1-IQ1_S123B23.13 GiB0.19 GiB23.90 GiB0.10 GiB28±37%
GLM-4.7-Flash-DerestrictedMoEI1-Q6_K31.2B22.92 GiB0.41 GiB23.90 GiB0.10 GiB21±37%
Huihui-GLM-4.7-Flash-abliteratedMoEI1-Q6_K31.2B22.92 GiB0.41 GiB23.90 GiB0.10 GiB21±37%
GLM-4.7-Flash-Claude-Opus-4.5-High-Reasoning-DistillMoEQ6_K31.2B22.92 GiB0.41 GiB23.90 GiB0.10 GiB21±37%
GLM-Z1-Rumination-32B-0414Q5_K_S33.1B21.35 GiB1.91 GiB23.89 GiB0.11 GiB6±8.3%
Maenad-70BI1-IQ2_S70.6B20.71 GiB2.50 GiB23.89 GiB0.11 GiB6±8.3%
DeepSeek-R1-Distill-Llama-70B-Uncensored-v2-Unbiased-ReasonerI1-IQ2_S70.6B20.71 GiB2.50 GiB23.89 GiB0.11 GiB6±8.3%
Rombos-LLM-70b-Llama-3.3I1-IQ2_S70.6B20.71 GiB2.50 GiB23.89 GiB0.11 GiB6±8.3%
L3.3-Electra-R1-70bI1-IQ2_S70.6B20.71 GiB2.50 GiB23.89 GiB0.11 GiB6±8.3%
L3.3-70B-Magnum-v4-SEIQ2_S70.6B20.71 GiB2.50 GiB23.89 GiB0.11 GiB6±8.3%
Latxa-Llama-3.1-70B-Instruct-v2I1-IQ2_S70.6B20.71 GiB2.50 GiB23.89 GiB0.11 GiB6±8.3%
Llama-3.3_70_b_uncensored_continuedI1-IQ2_S70.6B20.71 GiB2.50 GiB23.89 GiB0.11 GiB6±8.3%
Llama-3.3-70B-Instruct-abliteratedI1-IQ2_S70.6B20.71 GiB2.50 GiB23.89 GiB0.11 GiB6±8.3%
grok-oss-Revenant-70BI1-IQ2_S70.6B20.71 GiB2.50 GiB23.89 GiB0.11 GiB6±8.3%
Llama-3.1-Nemotron-70B-Instruct-HFI1-IQ2_S70.6B20.71 GiB2.50 GiB23.89 GiB0.11 GiB6±8.3%
L3.3-70B-Euryale-v2.3I1-IQ2_S70.6B20.71 GiB2.50 GiB23.89 GiB0.11 GiB6±8.3%
Hermes-3-Llama-3.1-70BIQ2_S70.6B20.71 GiB2.50 GiB23.89 GiB0.11 GiB6±8.3%
Hermes-4-70B-hereticI1-IQ2_S70.6B20.71 GiB2.50 GiB23.89 GiB0.11 GiB6±8.3%
Llama-3.3-70B-InstructIQ2_S70.6B20.71 GiB2.50 GiB23.89 GiB0.11 GiB6±8.3%
Llama-3.1-70BIQ2_S70.6B20.71 GiB2.50 GiB23.89 GiB0.11 GiB6±8.3%
Anubis-70B-v1.2IQ2_S70.6B20.71 GiB2.50 GiB23.89 GiB0.11 GiB6±8.3%
Hermes-4-70BIQ2_S70.6B20.71 GiB2.50 GiB23.89 GiB0.11 GiB6±8.3%
Golem-70B-v1bI1-IQ2_S70.6B20.71 GiB2.50 GiB23.89 GiB0.11 GiB6±8.3%
DeepSeek-R1-Distill-Llama-70B-abliteratedI1-IQ2_S70.6B20.71 GiB2.50 GiB23.89 GiB0.11 GiB6±8.3%
DeepSeek-R1-Distill-Llama-70B-hereticI1-IQ2_S70.6B20.71 GiB2.50 GiB23.89 GiB0.11 GiB6±8.3%
DeepSeek-R1-Distill-Llama-70BIQ2_S70.6B20.71 GiB2.50 GiB23.89 GiB0.11 GiB6±8.3%
Legion-V2.1-LLaMa-70BI1-IQ2_S70.6B20.71 GiB2.50 GiB23.89 GiB0.11 GiB6±8.3%
Assistant_Pepe_70BI1-IQ2_S70.6B20.71 GiB2.50 GiB23.89 GiB0.11 GiB6±8.3%
SEMIKONG-70BIQ2_S70.6B20.71 GiB2.50 GiB23.89 GiB0.11 GiB6±8.3%
Infinity-Instruct-7M-Gen-Llama3_1-70BI1-IQ2_S70.6B20.71 GiB2.50 GiB23.89 GiB0.11 GiB6±8.3%
New-Dawn-Llama-3-70B-32K-v1.0I1-IQ2_S70.6B20.71 GiB2.50 GiB23.89 GiB0.11 GiB6±8.3%
Meta-Llama-3-70B-Instruct-abliterated-v3.5I1-IQ2_S70.6B20.71 GiB2.50 GiB23.89 GiB0.11 GiB6±8.3%
Athene-70BIQ2_S70.6B20.71 GiB2.50 GiB23.89 GiB0.11 GiB6±8.3%
L3.3-70B-Magnum-DiamondIQ2_S70.6B20.71 GiB2.50 GiB23.89 GiB0.11 GiB6±8.3%
Meta-Llama-3-70B-InstructIQ2_S70.6B20.71 GiB2.50 GiB23.89 GiB0.11 GiB6±8.3%
EXAONE-4.5-33BI1-Q5_K_M34.4B21.87 GiB1.34 GiB23.86 GiB0.14 GiB6±8.3%
c4ai-command-r-08-2024Q5_K_L32.3B21.94 GiB1.25 GiB23.86 GiB0.14 GiB6±8.3%
Seed-OSS-36B-Instruct-biprojected-norm-preserving-abliteratedI1-Q4_136.2B21.20 GiB2.00 GiB23.85 GiB0.15 GiB6±8.3%
Seed-OSS-36B-InstructQ4_136.2B21.20 GiB2.00 GiB23.85 GiB0.15 GiB6±8.3%
Hermes-4.3-36B-hereticI1-Q4_136.2B21.20 GiB2.00 GiB23.85 GiB0.15 GiB6±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 M5 run?
1969 of 2118 indexed open-weight models fit a Apple M5 at 8,192 context with f16 KV cache, the largest being Gemma-4-Novelist-Eclipse-31B at Q5_K_S. That covers text, vision-language, image, video and speech models.
How much usable memory does a Apple M5 actually have?
Its nameplate is 32 GB, but about 22.32 GiB is available to a model once driver and compositor overhead is accounted for, and only 24 GB of the pool can be allocated to the GPU at all.
Is a Apple M5 fast for local AI?
Its memory bandwidth is 154 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.