What is Mercury 3?
Inception Labs announced Mercury 3, a new language model, on its blog. The post is titled "Diffusion at scale," suggesting the company aims to apply its diffusion-based language modeling approach at a larger scale.
Why diffusion matters
Most language models generate text token by token, left to right. Diffusion models adapt a technique used in image generation to text, progressively denoising an output. In theory, this promises faster and more parallel generation.
Details remain limited
The public source does not include the model's parameter count, context length, access terms or benchmark results, so a comparison with rivals is not yet possible.
The announcement drew attention on Hacker News with 163 points.
