AI ready MacBooks

We mark a MacBook as AI ready when it has Apple silicon and Apple lists a unified memory option of 32GB or more for that model.

On Apple silicon, the CPU, GPU and Neural Engine share one pool of unified memory. A language model run on the Mac itself (a local LLM) has to fit its weights in that pool, with room left for its working memory (the KV cache, which grows with context length), for macOS and for other apps.

Memory is usually the limit for local models, so we use memory capacity as the criterion. With 32GB, a model that is quantized to around 4 bits and has up to roughly 30 billion parameters can typically be loaded with room for a working context. More memory allows larger models or longer contexts. Speed also depends on the chip's GPU core count and memory bandwidth, which differ between M-series generations and tiers.

The figure is the highest memory option Apple lists for the model. Memory cannot be upgraded after purchase, so a used unit has whatever it was ordered with. Check the memory stated in the listing; many units were sold with 8GB to 24GB.

AI ready is our own filter label, not an Apple certification or a guarantee that any particular model or app will run. Intel MacBooks are not included: their memory is not shared with the GPU in the same way.

Definition used by the filter: chip_family in M1-M5 and ram_gb_max >= 32. Data: /m/ai-ready/ (JSON · Markdown).