Apple Silicon unified-memory guide
Local AI on 8GB Apple Silicon unified memory
Apple Silicon uses unified memory with Metal-capable local inference paths when the selected runtime supports them. AIPCFit reserves 3GB for macOS, apps, and runtime overhead, leaving about 5GB as the local model planning budget.
This page is for Apple Silicon Macs only. Intel Macs and incomplete Mac profiles are handled conservatively by the fit engine and do not receive Apple Silicon recommendations.
Catalog variants within this memory budget
Good fit
Llama 3.2 3B
Llama 3.2 3B · Q4_K_M · 3.7 GB planning memory · 2.7 GB minimum load
Good fit
Phi-3.5 Mini
Phi-3.5 Mini · Q4_K_M · 4.2 GB planning memory · 3.2 GB minimum load
Good fit
Phi-4 Mini
Phi-4 Mini · Q4_K_M · 4.2 GB planning memory · 3.2 GB minimum load
Good fit
Gemma 3 4B
Gemma 3 4B · Q4_K_M · 4.5 GB planning memory · 3.5 GB minimum load
Good fit
Qwen3 4B
Qwen3 4B · Q4_K_M · 4.5 GB planning memory · 3.5 GB minimum load
Fits
Llama 3.2 3B
Llama 3.2 3B · Q8_0 · 5.2 GB planning memory · 4.2 GB minimum load
Tight
DeepSeek-R1 Distill Qwen 7B
DeepSeek-R1 Distill Qwen 7B · Q4_K_M · 6.1 GB planning memory · 5.1 GB minimum load
Tight
Mistral 7B
Mistral 7B · Q4_K_M · 6.1 GB planning memory · 5.1 GB minimum load
Tight
Gemma 3 4B
Gemma 3 4B · Q8_0 · 6.2 GB planning memory · 5.2 GB minimum load
Tight
Qwen3 4B
Qwen3 4B · Q8_0 · 6.2 GB planning memory · 5.2 GB minimum load
Tight
Llama 3.1 8B
Llama 3.1 8B · Q4_K_M · 6.6 GB planning memory · 5.6 GB minimum load