Apple’s New AI Model Writes Long Texts Up to 128x Faster

Apple researchers have developed a powerful new language model that can write long passages incredibly fast — up to 128 times quicker than current AI systems like ChatGPT.

How Current AI Models Work

Most AI chatbots, including ChatGPT, use what’s called an autoregressive model.
That means they write text word by word, checking each new word against the ones already written. It’s accurate but slow.

How Apple’s Model Is Different

Apple’s new approach uses diffusion models, which generate many words in parallel and then refine the text in a few quick steps.
A special version of this, called a flow-matching model, skips most of those refinement steps and instantly produces the final text.

Meet FS-DFM

Apple’s study, titled “FS-DFM: Fast and Accurate Long Text Generation with Few-Step Diffusion Language Models,” introduces a new system called Few-Step Discrete Flow-Matching (FS-DFM).

Instead of taking hundreds or thousands of steps to write something coherent, FS-DFM can do it in just eight rounds — without losing quality.

Why It Matters

Apple and Ohio State University researchers tested FS-DFM and found:

• It matches or beats much larger AI models in text quality.
• It produces natural, confident writing with fewer errors.
• It’s far more efficient — smaller models can perform like big ones.

What’s Next

Apple plans to release the code and model checkpoints so other researchers can build on this work.

This could mark a big leap forward for AI text generation — making it faster, smarter, and more energy-efficient.