Explore AI
Mac with 32 GB
The comfortable tier. This is where 30B-class models become practical.
The comfortable tier. This is where 30B-class models become practical.
- Platform
- Apple Silicon
- Memory
- 32 GB unified
32 GB of unified memory is unusually good value for local inference, because the GPU can address all of it. A 30B model at Q4 with a long context fits.
Runs comfortably
- Qwen3 32B 24 GB at Q4 Apache-2.0
- Gemma 4 31B 24 GB at Q4 Apache-2.0
- Qwen3 8B 8 GB at Q4 Apache-2.0
- Qwen3 Coder 30B A3B 20 GB at Q4 · 30B total, 3B active Apache-2.0
Will not fit
- GPT-OSS 120Bneeds about 64 GB
Worth knowing
Diminishing returns start here for chat. The gains above this tier are mostly about running larger models for coding and agents.