16GB vs 32GB vs 36GB: why the line is 36GB
32GB feels like it should be close enough. For a stated 36GB minimum, close does not count - here is the honest maths and what each tier can and cannot do.
The three tiers against the requirement
16GB: Half the requirement, with change. Comfortable for everyday computing and cloud AI in a browser, but nowhere near the announced local-mode floor.
32GB: The frustrating tier. It is a lot of memory for most tasks, and only 4GB short - but a minimum is a minimum. On the announced requirements, a 32GB Mac does not run Perplexity Computer local mode.
36GB: The floor itself. Configurations at exactly 36GB meet the requirement. The next common steps up - 48GB and 64GB - meet it with headroom.
Why local AI eats memory
When a model runs locally, the model itself, its context and the agent's working data all occupy unified memory, which the CPU and GPU share. There is no separate pool to borrow from. That is why vendors set a hard floor rather than a vague recommendation, and why the floor sits above the memory most laptops ship with.
Should you buy more than 36GB?
Headroom helps when you multitask alongside a local model, and memory cannot be added later - so buyers who keep machines for many years often step to 48GB or 64GB. That is a judgement call about your budget and habits, not a requirement. We make no benchmark claims for any tier.
Frequently asked questions
Why does 32GB not count if it is only 4GB short?
Because 36GB was announced as the minimum requirement. Software gates are binary: below the minimum does not meet it, however close it is.
Is 36GB enough, or do I need 64GB?
36GB meets the announced requirement. More memory is headroom for multitasking and longevity, not an eligibility requirement.
Can external RAM or an eGPU help?
No. Apple Silicon Macs use fixed unified memory, and the requirement is about that unified memory pool.