Inception Drops Mercury 2: A Diffusion Model Built for Speed
Stefano Ermon's AI startup Inception launches Mercury 2, a diffusion-based model promising faster and cheaper responses than competitors.
Inception, the AI company founded by Stanford professor Stefano Ermon, just dropped Mercury 2 — a diffusion-based AI model that takes a fundamentally different approach to answering user questions.
The pitch is simple: significantly faster and significantly cheaper than the competition.
Ermon is no newcomer to the diffusion game. He helped pioneer the core diffusion technology that now powers some of the most popular AI image and video generation tools on the market. Mercury 2 applies that same underlying approach to conversational AI.
While most major chatbots rely on autoregressive transformer architectures — generating text one token at a time — diffusion models work differently, potentially unlocking serious speed advantages.
If the performance claims hold up, Mercury 2 could put real pressure on incumbents racing to cut inference costs. Speed and price are increasingly where the AI wars are being fought.