Apple unveiled a laptop built around its new M6 chip on 27 October, the 14-inch MacBook Pro, and the pitch leans hard on running language models on the laptop itself. Pre-orders open on 28 October, with deliveries from 6 November, at a starting price of $1,999 in the U.S. The speed claims below all come from Apple’s press release, in the company’s own words and on its own tests.
The headline number is for the wait before a model starts answering. Apple says time to first token, the delay between sending a prompt and seeing the first word of the reply, is up to 13x faster than on the older M1 13-inch model. Against the 14-inch M5 machine the gain shrinks to 1.3x. Apple also puts the GPU at up to 25 percent faster for AI workloads such as prompt processing compared with M5, and says its Neural Engine runs up to 1.9x faster than M5’s.
Read closely, those figures describe how quickly a model digests your input. The release does not say how fast the machine produces a long answer once it starts, which is the number people running a local chatbot feel most. It also names no model, no model size, and no quantization level behind the 13x result, so nobody can reproduce it yet. Apple’s footnotes say the M5 comparison machine had 32GB of memory and a 4TB drive, a more generous setup than most buyers will choose.
Memory is where local AI usually runs out of road. The M6 model ships with 16GB of unified memory and can be configured to 32GB, with bandwidth up to 170GB/s. Apple presents that bandwidth as a help for demanding apps and large projects. For language models it matters twice, because the weights have to fit in memory and the chip has to read them quickly. A 32GB ceiling caps the size of model you can load, however fast the processor is.
Apple also frames the chip around privacy, saying the extra accelerators inside the GPU, plus the Neural Engine, let users run LLMs locally “all while preserving user privacy.” That is a real argument for lawyers, doctors, and anyone whose documents cannot leave the building. It sits next to a less comfortable footnote: Apple says certain Apple Intelligence features run on server-side models and carry daily usage limits, with wider access to come as a paid option. Local processing and cloud features are separate things in this product, and the privacy line only covers the first.
The upgrade story is aimed at older Macs. Apple says anyone upgrading from M1 sees AI performance up to 13x faster and graphics up to 11x faster, plus a 16GB starting memory and a battery rated for as long as 24 hours. For an M1 owner who has found local models sluggish, that is a large jump on paper. If you already own an M5, the release itself offers 1.3x and 25 percent as the case for spending $1,999.
Developers have an extra reason to look. Apple lists up to 2.5x faster Xcode builds than the M1 machine and 1.3x faster than M5, which matters if the same laptop compiles your app and hosts the model it calls.
Apple published the claims in its Newsroom press release on 27 October 2026. Independent benchmarks should arrive once reviewers get units in November, and the useful test is a long generation run with a model that nearly fills memory, not a prompt-processing chart. Until then, treat the 13x figure as a ceiling Apple measured under its own conditions, and size your memory order for the model you want to run, not for the speedup you were promised.
Reported by Apple Newsroom on 27 October 2026.
