Extreme unified-memory class. Roughly 200B at 4-bit is tight; 405B does not fit realistically. Previous generation.
Local AI fit
Comfortable up to roughly 120B. About 200B Q4 is tight with short context; 405B requires a larger memory tier.
200B is a capacity edge, not a blanket promise: it generally requires strong weight quantization, a short context, and little parallel load. KV-cache precision and the model architecture can materially change the result.
192GB explained Why unified memory is more than model weights Understand quantization, KV cache, macOS headroom, and the difference between “fits” and “runs well.” Open RAM guidePurchase status
Most compelling when priced clearly below a current model with similar memory.
At a glance
What works well
What to consider
Data verified
Product data and links last checked on 2026-06-03. Prices and availability can change at any time.
Compare before buying
M5-Max128 GB Unified Memory
M5 Max with 128GB unified memory and 614GB/s memory bandwidth; currently available new.
Last checked: 2026-07-28
M4-Max128 GB Unified Memory
M4 Max with 128GB unified memory and 546GB/s memory bandwidth; currently available new.
Last checked: 2026-07-28
M1-Ultra128 GB Unified Memory Refurbished only
M1 Ultra with 128GB unified memory and 800GB/s memory bandwidth; primarily available used or refurbished.
Last checked: 2026-07-28
M2-Ultra64 GB Unified Memory24-Core CPU60-Core GPU Previous generation
Classic for 70B in 4-bit. Previous generation — cheaper than current alternatives.
Last checked: 2026-06-03