Apple · apple
Apple M4
Apple M4 has 16 GB of unified memory at 120 GB/s — about 11.16 GiB usable after driver and compositor overhead. 748 of 2118 indexed models fit at 128K context with f16 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
LPDDR5X-7500
Bandwidth
120 GB/s
128-bit bus
Tensor FP16
—
dense
TDP
—
text 581video 15vision language 85audio asr 33embedding 15audio tts 18image 1
What fits at 128K context
largest quantization that fits, per model · 748 of 2118 indexed
| Model | Best quant | Params | Weights● | KV● | Total in memory◐ | Headroom◐ | tok/s◐ |
|---|---|---|---|---|---|---|---|
| Yi-6B-Chat | I1-Q4_K_M | 6.1B | 3.42 GiB | 8.00 GiB | 12.00 GiB | 0.00 GiB | 9±8.3% |
| Yi-1.5-6B-Chat | Q4_K_M | 6.1B | 3.42 GiB | 8.00 GiB | 12.00 GiB | 0.00 GiB | 9±8.3% |
| Wan2.1-VACE-14B | Q5_K_S | 17.3B | 11.41 GiB | 0.00 GiB | 12.00 GiB | 0.00 GiB | 9±8.3% |
| YuYu1015-Ornith-1.0-9B-abliterated | UD-Q6_K | 9.4B | 7.40 GiB | 4.00 GiB | 11.99 GiB | 0.01 GiB | 9±8.3% |
| granite-3.3-2b-instruct | Q4_K_M | 2.5B | 1.44 GiB | 10.00 GiB | 11.99 GiB | 0.01 GiB | 9±8.3% |
| granite-3.2-2b-instruct | Q4_K_M | 2.5B | 1.44 GiB | 10.00 GiB | 11.99 GiB | 0.01 GiB | 9±8.3% |
| granite-3.1-2b-instruct | Q4_K_M | 2.5B | 1.44 GiB | 10.00 GiB | 11.99 GiB | 0.01 GiB | 9±8.3% |
| granite-vision-3.2-2b | Q4_K_M | 3.0B | 1.44 GiB | 10.00 GiB | 11.99 GiB | 0.01 GiB | 9±8.3% |
| granite-vision-3.3-2b | Q4_K_M | 3.0B | 1.44 GiB | 10.00 GiB | 11.99 GiB | 0.01 GiB | 9±8.3% |
| granite-speech-4.1-2b-nar | Q4_K_M | 2.3B | 1.45 GiB | 10.00 GiB | 11.99 GiB | 0.01 GiB | 9±8.3% |
| GrammarCoder-7B-Base | I1-Q4_K_M | 7.6B | 4.37 GiB | 7.00 GiB | 11.98 GiB | 0.02 GiB | 9±8.3% |
| SmolLM3-3B | Q6_K_L | 3.1B | 2.42 GiB | 9.00 GiB | 11.98 GiB | 0.02 GiB | 9±8.3% |
| InternVL3_5-30B-A3B | IQ3_XXS | 30.8B | 11.38 GiB | 0.00 GiB | 11.97 GiB | 0.03 GiB | 9±8.3% |
| Teuken-7B-instruct-research-v0.4 | Q8_0 | 7.5B | 7.38 GiB | 4.00 GiB | 11.97 GiB | 0.03 GiB | 9±8.3% |
| Qwen3.6-28BMoE | I1-IQ2_M | 28.2B | 8.91 GiB | 2.50 GiB | 11.97 GiB | 0.03 GiB | 16±37% |
| Qwen3.5-28BMoE | I1-IQ2_M | 28.7B | 8.91 GiB | 2.50 GiB | 11.97 GiB | 0.03 GiB | 16±37% |
| DeepHat-V1-7B-Heretic-Abliterated | I1-Q4_K_M | 7.6B | 4.36 GiB | 7.00 GiB | 11.97 GiB | 0.03 GiB | 9±8.3% |
| ShizhenGPT-7B-VL | I1-Q4_K_M | 8.3B | 4.36 GiB | 7.00 GiB | 11.97 GiB | 0.03 GiB | 9±8.3% |
| DeepHat-V1-7B | Q4_K_M | 7.6B | 4.36 GiB | 7.00 GiB | 11.97 GiB | 0.03 GiB | 9±8.3% |
| HuatuoGPT-o1-7B | I1-Q4_K_M | 7.6B | 4.36 GiB | 7.00 GiB | 11.97 GiB | 0.03 GiB | 9±8.3% |
| MathSmith-DS-Qwen-7B-LongCoT | I1-Q4_K_M | 7.6B | 4.36 GiB | 7.00 GiB | 11.97 GiB | 0.03 GiB | 9±8.3% |
| AstraGPTCoder-7B | I1-Q4_K_M | 7.6B | 4.36 GiB | 7.00 GiB | 11.97 GiB | 0.03 GiB | 9±8.3% |
| Qwen2.5-Coder-7B-Instruct-Ghidra-v2 | I1-Q4_K_M | 7.6B | 4.36 GiB | 7.00 GiB | 11.97 GiB | 0.03 GiB | 9±8.3% |
| EsDrac-v1-7B | I1-Q4_K_M | 7.6B | 4.36 GiB | 7.00 GiB | 11.97 GiB | 0.03 GiB | 9±8.3% |
| Hemlock-Apothecary-7B-GRPO-e3 | I1-Q4_K_M | 7.6B | 4.36 GiB | 7.00 GiB | 11.97 GiB | 0.03 GiB | 9±8.3% |
| openhands-lm-7b-v0.1 | I1-Q4_K_M | 7.6B | 4.36 GiB | 7.00 GiB | 11.97 GiB | 0.03 GiB | 9±8.3% |
| Hemlock2-Coder-7B-GRPO | I1-Q4_K_M | 7.6B | 4.36 GiB | 7.00 GiB | 11.97 GiB | 0.03 GiB | 9±8.3% |
| shellwhiz-7b | I1-Q4_K_M | 7.6B | 4.36 GiB | 7.00 GiB | 11.97 GiB | 0.03 GiB | 9±8.3% |
| Qwen2.5-Coder-7B-Instruct-abliterated | I1-Q4_K_M | 7.6B | 4.36 GiB | 7.00 GiB | 11.97 GiB | 0.03 GiB | 9±8.3% |
| Qwen2.5-Coder-7B-Instruct-OBLITERATED-advanced | I1-Q4_K_M | 7.6B | 4.36 GiB | 7.00 GiB | 11.97 GiB | 0.03 GiB | 9±8.3% |
| Qwen-STEM-Specialist-7B | I1-Q4_K_M | 7.6B | 4.36 GiB | 7.00 GiB | 11.97 GiB | 0.03 GiB | 9±8.3% |
| VulnLLM-R-7B | I1-Q4_K_M | 7.6B | 4.36 GiB | 7.00 GiB | 11.97 GiB | 0.03 GiB | 9±8.3% |
| Garnet-OCR-7B-0422 | I1-Q4_K_M | 8.3B | 4.36 GiB | 7.00 GiB | 11.97 GiB | 0.03 GiB | 9±8.3% |
| UwU-7B-Instruct | I1-Q4_K_M | 7.6B | 4.36 GiB | 7.00 GiB | 11.97 GiB | 0.03 GiB | 9±8.3% |
| Video-R1-7B | I1-Q4_K_M | 8.3B | 4.36 GiB | 7.00 GiB | 11.97 GiB | 0.03 GiB | 9±8.3% |
| HARC-Qwen2.5-7B-Instruct | I1-Q4_K_M | 7.6B | 4.36 GiB | 7.00 GiB | 11.97 GiB | 0.03 GiB | 9±8.3% |
| Qwen2.5-Coder-7B-Abliterated | I1-Q4_K_M | 7.6B | 4.36 GiB | 7.00 GiB | 11.97 GiB | 0.03 GiB | 9±8.3% |
| Bozdogan-7B | I1-Q4_K_M | 7.6B | 4.36 GiB | 7.00 GiB | 11.97 GiB | 0.03 GiB | 9±8.3% |
| Qwen2.5-7B-Instruct-abliterated-v2 | Q4_K_M | 7.6B | 4.36 GiB | 7.00 GiB | 11.97 GiB | 0.03 GiB | 9±8.3% |
| Crazy-AI-Model | I1-Q4_K_M | 7.6B | 4.36 GiB | 7.00 GiB | 11.97 GiB | 0.03 GiB | 9±8.3% |
| turbo-ai-7b | I1-Q4_K_M | 7.6B | 4.36 GiB | 7.00 GiB | 11.97 GiB | 0.03 GiB | 9±8.3% |
| DeepSeek-R1-Distill-Qwen-7B-abliterated-v2 | I1-Q4_K_M | 7.6B | 4.36 GiB | 7.00 GiB | 11.97 GiB | 0.03 GiB | 9±8.3% |
| Ghosty-7B | I1-Q4_K_M | 7.6B | 4.36 GiB | 7.00 GiB | 11.97 GiB | 0.03 GiB | 9±8.3% |
| Bernini-MLLM-Qwen2.5-VL-7B | Q4_K_M | 8.3B | 4.36 GiB | 7.00 GiB | 11.97 GiB | 0.03 GiB | 9±8.3% |
| Qwen2.5-Math-7B-Instruct | Q4_K_M | 7.6B | 4.36 GiB | 7.00 GiB | 11.97 GiB | 0.03 GiB | 9±8.3% |
| Qwen3-7B-Instruct | Q4_K_M | 7.6B | 4.36 GiB | 7.00 GiB | 11.97 GiB | 0.03 GiB | 9±8.3% |
| SP-7B | I1-Q4_K_M | 7.6B | 4.36 GiB | 7.00 GiB | 11.97 GiB | 0.03 GiB | 9±8.3% |
| Qwen2.5-7B-Instruct | Q4_K_M | 7.6B | 4.36 GiB | 7.00 GiB | 11.97 GiB | 0.03 GiB | 9±8.3% |
| Qwen2.5-Coder-7B-Instruct-Uncensored | I1-Q4_K_M | 7.6B | 4.36 GiB | 7.00 GiB | 11.97 GiB | 0.03 GiB | 9±8.3% |
| Qwen2.5-VL-7B-Instruct-abliterated | I1-Q4_K_M | 8.3B | 4.36 GiB | 7.00 GiB | 11.97 GiB | 0.03 GiB | 9±8.3% |
| DeepSeek-R1-Distill-Qwen-8B-Abliterated | I1-Q4_K_M | 7.6B | 4.36 GiB | 7.00 GiB | 11.97 GiB | 0.03 GiB | 9±8.3% |
| Qwen2.5-7B | Q4_K_M | 7.6B | 4.36 GiB | 7.00 GiB | 11.97 GiB | 0.03 GiB | 9±8.3% |
| OREAL-DeepSeek-R1-Distill-Qwen-7B | Q4_K_M | 7.6B | 4.36 GiB | 7.00 GiB | 11.97 GiB | 0.03 GiB | 9±8.3% |
| Qwen2.5-7B-Instruct-1M | Q4_K_M | 7.6B | 4.36 GiB | 7.00 GiB | 11.97 GiB | 0.03 GiB | 9±8.3% |
| LocalAI-functioncall-qwen2.5-7b-v0.5 | Q4_K_M | 7.6B | 4.36 GiB | 7.00 GiB | 11.97 GiB | 0.03 GiB | 9±8.3% |
| Qwen2.5-7B-Instruct-1M-abliterated | Q4_K_M | 7.6B | 4.36 GiB | 7.00 GiB | 11.97 GiB | 0.03 GiB | 9±8.3% |
| Qwen2.5-Coder-7B | Q4_K_M | 7.6B | 4.36 GiB | 7.00 GiB | 11.97 GiB | 0.03 GiB | 9±8.3% |
| UI-TARS-1.5-7B | Q4_K_M | 8.3B | 4.36 GiB | 7.00 GiB | 11.97 GiB | 0.03 GiB | 9±8.3% |
| Qwen2.5-Coder-7B-Instruct-Uncensored | Q4_K_M | 7.6B | 4.36 GiB | 7.00 GiB | 11.97 GiB | 0.03 GiB | 9±8.3% |
| DeepSeek-R1-Distill-Qwen-7B | Q4_K_M | 7.6B | 4.36 GiB | 7.00 GiB | 11.97 GiB | 0.03 GiB | 9±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 M4 run?
- 748 of 2118 indexed open-weight models fit a Apple M4 at 131,072 context with f16 KV cache, the largest being Yi-6B-Chat at I1-Q4_K_M. That covers text, vision-language, image, video and speech models.
- How much usable memory does a Apple M4 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 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.