Skip to content

Search stories, models, companies…Subscribe

Topic · Concept

Diffusion models

Diffusion models generate output by starting from random noise and refining it over several passes, instead of writing it one piece at a time. This page covers how they work, the papers behind them and the newer diffusion language models, such as Inception's Mercury 3, that apply the method to text.

1 story · updated Oct 20

Follow via RSS Get Diffusion models alerts

Key facts

Checked Oct 23

Original diffusion paper: "Deep Unsupervised Learning using Nonequilibrium Thermodynamics," Sohl-Dickstein, Weiss, Maheswaranathan and Ganguli; submitted March 12, 2015
DDPM: "Denoising Diffusion Probabilistic Models," Ho, Jain and Abbeel; submitted June 19, 2020
Latent diffusion: "High-Resolution Image Synthesis with Latent Diffusion Models," Rombach et al.; submitted December 20, 2021
LLaDA: "Large Language Diffusion Models," Nie et al.; submitted February 14, 2025; an 8B diffusion language model
Gemini Diffusion: Announced May 20, 2025, as an experimental text diffusion model from Google DeepMind; demo access by waitlist
Mercury paper: "Mercury: Ultra-Fast Language Models Based on Diffusion," Inception Labs et al.; submitted June 17, 2025
Mercury 3 launch: October 19, 2026; Inception calls it "the most capable diffusion-based LLM released to date"
Mercury 3 speed (vendor-reported): 1,068 tokens per second on Nvidia GPUs, per Inception
Mercury 3 context and price: 1M-token context; $0.20 per million input tokens and $0.70 per million output tokens list price, with a 50% launch discount

Start here

Models · ExplainerWhat is a diffusion language model? Inception's Mercury 3 explainedA diffusion language model drafts a whole block of text and refines it in parallel, instead of writing one token at a time. Inception's Mercury 3, launched October 19, 2026, claims 1,068 tokens per second, a 1-million-token context and $0.20/$0.70 pricing. Here's how diffusion LLMs work and what the trade-offs are.Oct 20·7 min read

Latest on Diffusion models

Models · ExplainerWhat is a diffusion language model? Inception's Mercury 3 explainedA diffusion language model drafts a whole block of text and refines it in parallel, instead of writing one token at a time. Inception's Mercury 3, launched October 19, 2026, claims 1,068 tokens per second, a 1-million-token context and $0.20/$0.70 pricing. Here's how diffusion LLMs work and what the trade-offs are.Oct 20·7 min read

Nothing in this format on this page — try another filter or the next page.