AI · 1h ago
Google and Inception Labs race to make AI write in parallel blocks
Google released DiffusionGemma, an open-weight diffusion text model, while Inception Labs launched Mercury 2 as a hosted service. Both generate entire paragraphs at once instead of word by word, promising faster output. The open model requires local setup, while the hosted one offers instant access but is closed.
Meridian48 take
The speed advantage is clear, but quality comparisons are still needed to determine if diffusion models can replace autoregressive ones in production.
Read the full reporting
Two labs race to make AI write whole paragraphs at once instead of word by word →
DEV Community
diffusion-text-modelsparallel-generation