Apple · apple
Apple M3
Apple M3 has 8 GB of unified memory at 102 GB/s — about 5.58 GiB usable after driver and compositor overhead. 732 of 2118 indexed models fit at 32K context with f16 KV. Note only 6 GB of its 8 GB is allocatable to the GPU.
Spec sheet· bandwidth, theoreticalFrom the file· fit from summed bytesPredicted· speed
Memory
8 GB
LPDDR5-6400
Bandwidth
102 GB/s
128-bit bus
Tensor FP16
—
dense
TDP
—
text 572vision language 78audio tts 19embedding 21audio asr 37video 5
What fits at 32K context
largest quantization that fits, per model · 732 of 2118 indexed
| Model | Best quant | Params | Weights● | KV● | Total in memory◐ | Headroom◐ | tok/s◐ |
|---|---|---|---|---|---|---|---|
| Qwen3.5-9B-Coder | I1-Q3_K_M | 9.7B | 4.41 GiB | 1.00 GiB | 6.00 GiB | 0.00 GiB | 15±8.3% |
| Qwopus3.5-9B-v3.5 | Q3_K_M | 9.7B | 4.41 GiB | 1.00 GiB | 6.00 GiB | 0.00 GiB | 15±8.3% |
| Qwythos-9B-Claude-Mythos-5-1M-MTP | I1-Q3_K_M | 9.7B | 4.41 GiB | 1.00 GiB | 6.00 GiB | 0.00 GiB | 15±8.3% |
| Huihui-Qwythos-9B-Claude-Mythos-5-1M-abliterated | I1-Q3_K_M | 9.7B | 4.41 GiB | 1.00 GiB | 6.00 GiB | 0.00 GiB | 15±8.3% |
| Qwen3.5-9B-Fable-5-v1 | I1-Q3_K_M | 9.7B | 4.41 GiB | 1.00 GiB | 6.00 GiB | 0.00 GiB | 15±8.3% |
| Qwythos-9B-v2 | I1-Q3_K_M | 9.7B | 4.41 GiB | 1.00 GiB | 6.00 GiB | 0.00 GiB | 15±8.3% |
| PINQWEN-3.5-9B-1M-BF16 | I1-Q3_K_M | 9.7B | 4.41 GiB | 1.00 GiB | 6.00 GiB | 0.00 GiB | 15±8.3% |
| Openprose-2-Flash | I1-Q3_K_M | 9.7B | 4.41 GiB | 1.00 GiB | 6.00 GiB | 0.00 GiB | 15±8.3% |
| Qwen3.5-9B-Nikusui-v1 | I1-Q3_K_M | 9.7B | 4.41 GiB | 1.00 GiB | 6.00 GiB | 0.00 GiB | 15±8.3% |
| Ornstein-3.5-9B-V1.5 | I1-Q3_K_M | 9.7B | 4.41 GiB | 1.00 GiB | 6.00 GiB | 0.00 GiB | 15±8.3% |
| Ornith-1.0-9B-heretic-MTP | I1-Q3_K_M | 9.4B | 4.41 GiB | 1.00 GiB | 6.00 GiB | 0.00 GiB | 15±8.3% |
| Tess-4-9B | I1-Q3_K_M | 9.7B | 4.41 GiB | 1.00 GiB | 6.00 GiB | 0.00 GiB | 15±8.3% |
| dotwebs-1 | I1-Q3_K_M | 9.7B | 4.41 GiB | 1.00 GiB | 6.00 GiB | 0.00 GiB | 15±8.3% |
| lift | Q3_K_M | 9.7B | 4.41 GiB | 1.00 GiB | 6.00 GiB | 0.00 GiB | 15±8.3% |
| Hemlock-Qwopus3.5-9B-Coder | I1-Q3_K_M | 9.7B | 4.41 GiB | 1.00 GiB | 6.00 GiB | 0.00 GiB | 15±8.3% |
| Qwen3.5-9B-DeepSeek-V4-Flash | Q3_K_M | 9.7B | 4.41 GiB | 1.00 GiB | 6.00 GiB | 0.00 GiB | 15±8.3% |
| Yi-6B-Chat | I1-Q4_K_M | 6.1B | 3.42 GiB | 2.00 GiB | 6.00 GiB | 0.00 GiB | 15±8.3% |
| Yi-1.5-6B-Chat | Q4_K_M | 6.1B | 3.42 GiB | 2.00 GiB | 6.00 GiB | 0.00 GiB | 15±8.3% |
| Qwen3.6-9B-Heretic-Uncensored-Thinking-Sweet-Madness | Q4_K_S | 9.1B | 4.88 GiB | 0.50 GiB | 5.99 GiB | 0.01 GiB | 15±8.3% |
| Phi-4-mini-reasoning | UD-IQ2_M | 3.8B | 1.43 GiB | 4.00 GiB | 5.99 GiB | 0.01 GiB | 15±8.3% |
| MARTHA-LXVII.8B_QWEN-3.5-3.6_prune_9b-3.6_base | I1-Q4_K_S | 8.1B | 4.52 GiB | 0.88 GiB | 5.99 GiB | 0.01 GiB | 15±8.3% |
| Fara1.5-9B | IQ3_M | 9.4B | 4.40 GiB | 1.00 GiB | 5.98 GiB | 0.02 GiB | 15±8.3% |
| QwenPaw-Flash-9B | IQ3_M | 9.4B | 4.40 GiB | 1.00 GiB | 5.98 GiB | 0.02 GiB | 15±8.3% |
| grug-9b | IQ3_M | 9.4B | 4.40 GiB | 1.00 GiB | 5.98 GiB | 0.02 GiB | 15±8.3% |
| OmniCoder-9B | IQ3_M | 9.4B | 4.40 GiB | 1.00 GiB | 5.98 GiB | 0.02 GiB | 15±8.3% |
| Ornith-1.0-9B | IQ3_M | 9.2B | 4.40 GiB | 1.00 GiB | 5.98 GiB | 0.02 GiB | 15±8.3% |
| Qwen3.5-9B-Neo | IQ3_M | 9.7B | 4.40 GiB | 1.00 GiB | 5.98 GiB | 0.02 GiB | 15±8.3% |
| Aura-4B | I1-IQ2_XS | 4.5B | 1.41 GiB | 4.00 GiB | 5.98 GiB | 0.02 GiB | 15±8.3% |
| magnum-v2-4b | I1-IQ2_XS | 4.5B | 1.41 GiB | 4.00 GiB | 5.98 GiB | 0.02 GiB | 15±8.3% |
| orpheus-3b-0.1-pretrained | Q2_K_L | 3.8B | 1.92 GiB | 3.50 GiB | 5.97 GiB | 0.03 GiB | 15±8.3% |
| Llama-3.2-3B-Instruct-uncensored | IQ4_XS | 3.6B | 1.91 GiB | 3.50 GiB | 5.97 GiB | 0.03 GiB | 15±8.3% |
| Teuken-7B-instruct-research-v0.4 | I1-Q4_K_S | 7.5B | 4.38 GiB | 1.00 GiB | 5.97 GiB | 0.03 GiB | 15±8.3% |
| Phi-4-mini-instruct | IQ2_M | 3.8B | 1.40 GiB | 4.00 GiB | 5.96 GiB | 0.04 GiB | 15±8.3% |
| Qwen3.6-12B-IQ-Ultra-Heretic-Uncensored-Thinking-V2-Hightop | Q2_K | 12.1B | 4.60 GiB | 0.75 GiB | 5.96 GiB | 0.04 GiB | 15±8.3% |
| glm-4v-9b | Q4_K_S | 13.9B | 5.36 GiB | 0.00 GiB | 5.96 GiB | 0.04 GiB | 15±8.3% |
| Llama-3.2-3B-Instruct-abliterated | I1-IQ4_XS | 3.6B | 1.90 GiB | 3.50 GiB | 5.96 GiB | 0.04 GiB | 15±8.3% |
| starcoder2-7bKV unresolved | Q3_K_M | 7.2B | 3.35 GiB | 2.00 GiB | 5.95 GiB | 0.05 GiB | 15±8.3% |
| InternVL3_5-14B | Q2_K | 15.1B | 5.36 GiB | 0.00 GiB | 5.95 GiB | 0.05 GiB | 15±8.3% |
| Gemma-4-E4B-Luchador | Q4_K_S | 8.0B | 4.88 GiB | 0.51 GiB | 5.95 GiB | 0.05 GiB | 15±8.3% |
| nomic-embed-code | Q3_K_L | 7.1B | 3.59 GiB | 1.75 GiB | 5.94 GiB | 0.06 GiB | 15±8.3% |
| ACE-Step-v1-3.5B | Q3_K_L | 3.3B | 5.35 GiB | 0.00 GiB | 5.94 GiB | 0.06 GiB | 15±8.3% |
| GLM-4.6V-Flash | IQ3_XS | 10.3B | 4.10 GiB | 1.25 GiB | 5.94 GiB | 0.06 GiB | 15±8.3% |
| glm4.1v-9b-base-sft | I1-IQ3_XS | 10.3B | 4.10 GiB | 1.25 GiB | 5.94 GiB | 0.06 GiB | 15±8.3% |
| GLM-Z1-9B-0414 | IQ3_XS | 9.4B | 4.10 GiB | 1.25 GiB | 5.94 GiB | 0.06 GiB | 15±8.3% |
| GLM-4-9B-0414 | IQ3_XS | 9.4B | 4.10 GiB | 1.25 GiB | 5.94 GiB | 0.06 GiB | 15±8.3% |
| Dolphin3.0-Llama3.2-3B | Q4_K_M | 3.2B | 1.88 GiB | 3.50 GiB | 5.94 GiB | 0.06 GiB | 15±8.3% |
| Llama-Doctor-3.2-3B-Instruct | I1-Q4_K_M | 3.2B | 1.88 GiB | 3.50 GiB | 5.94 GiB | 0.06 GiB | 15±8.3% |
| Llama-Song-Stream-3B-Instruct | Q4_K_M | 3.2B | 1.88 GiB | 3.50 GiB | 5.94 GiB | 0.06 GiB | 15±8.3% |
| Llama-3.2-3B-Instruct-roleplay-tuned | I1-Q4_K_M | 3.2B | 1.88 GiB | 3.50 GiB | 5.94 GiB | 0.06 GiB | 15±8.3% |
| Llama-3.2-3B-Instruct-heretic-ablitered-uncensored | I1-Q4_K_M | 3.2B | 1.88 GiB | 3.50 GiB | 5.94 GiB | 0.06 GiB | 15±8.3% |
| llama-3.2-Korean-Bllossom-3B | Q4_K_M | 3.2B | 1.88 GiB | 3.50 GiB | 5.94 GiB | 0.06 GiB | 15±8.3% |
| Llama-3.2-3B-Instruct-uncensored | Q4_K_M | 3.2B | 1.88 GiB | 3.50 GiB | 5.94 GiB | 0.06 GiB | 15±8.3% |
| llama-3.2-3b-instruct-bnb-4bit | Q4_K_M | 3.3B | 1.88 GiB | 3.50 GiB | 5.94 GiB | 0.06 GiB | 15±8.3% |
| Llama-3.2-3B-Instruct | Q4_K_M | 3.2B | 1.88 GiB | 3.50 GiB | 5.94 GiB | 0.06 GiB | 15±8.3% |
| Llama-3.2-3B | Q4_K_M | 3.2B | 1.88 GiB | 3.50 GiB | 5.94 GiB | 0.06 GiB | 15±8.3% |
| OneLLM-Doey-ChatQA-V1-Llama-3.2-3B | Q4_K_M | 3.2B | 1.88 GiB | 3.50 GiB | 5.94 GiB | 0.06 GiB | 15±8.3% |
| Llama3.2-3B-creative-writer-v0.1 | I1-Q4_K_M | 3.2B | 1.88 GiB | 3.50 GiB | 5.94 GiB | 0.06 GiB | 15±8.3% |
| Firefly-V3.2 | I1-Q4_K_M | 3.2B | 1.88 GiB | 3.50 GiB | 5.94 GiB | 0.06 GiB | 15±8.3% |
| Firefly-V3 | I1-Q4_K_M | 3.2B | 1.88 GiB | 3.50 GiB | 5.94 GiB | 0.06 GiB | 15±8.3% |
| Hermes-3-Llama-3.2-3B | Q4_K_M | 3.2B | 1.88 GiB | 3.50 GiB | 5.94 GiB | 0.06 GiB | 15±8.3% |
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 M3 run?
- 732 of 2118 indexed open-weight models fit a Apple M3 at 32,768 context with f16 KV cache, the largest being Qwen3.5-9B-Coder at I1-Q3_K_M. That covers text, vision-language, image, video and speech models.
- How much usable memory does a Apple M3 actually have?
- Its nameplate is 8 GB, but about 5.58 GiB is available to a model once driver and compositor overhead is accounted for, and only 6 GB of the pool can be allocated to the GPU at all.
- Is a Apple M3 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.