20 October 2026

Inception launches Mercury 3, a fast diffusion-based language model

• Inception, a company building diffusion language models, says its new model Mercury 3 is its most capable yet.
• Diffusion language models generate text differently from the usual chatbot approach, and Inception claims Mercury 3 matches leading models in quality.
• Inception says Mercury 3 produces more than 1,000 tokens per second, where a token is roughly a word fragment, so output is very fast.

How it was covered

Latent Spaceswyx & Alessio