Fetching from the wire…
01
02
03
04
05
06
07
08
Research2026-03-22 · source-backed
Think in Diffusion, Talk in Autoregression: drafts tokens in parallelized diffusion then outputs autoregressively in a single forward pass. The draft model is the base model itself. First architecture to close the quality gap with pure AR at nearly 6x throughput. Source
Each link below shares sources, entities, or timing with this story.
Shared entity: Diffusion / Same source domain / Shared topic / What happened next
Both cover Diffusion; reported by the same outlet (arxiv.org); overlapping topics (diffusion, model).
Shared entity: Speed / Same source domain / What happened next / Tension
Both cover Speed; reported by the same outlet (arxiv.org); picks up the Speed thread on 2026-04-22.
Shared entity: Diffusion / Shared topic / What happened next
Both cover Diffusion; overlapping topics (diffusion, model, quality); picks up the Diffusion thread on 2026-06-11.
Same source domain / Shared topic / Tension
Reported by the same outlet (arxiv.org); overlapping topics (diffusion, draft, model); pushes against this story (but).
Shared entity: Speed / Same source domain / What happened next
Both cover Speed; reported by the same outlet (arxiv.org); picks up the Speed thread on 2026-05-26.
Same source domain / Shared topic / Downstream implication
Reported by the same outlet (arxiv.org); overlapping topics (architecture, model); traces where this leads (which means).
Reported by the same outlet (arxiv.org); overlapping topics (architecture, model); traces where this leads (downstream).
Same source domain / Shared topic / Tension
Reported by the same outlet (arxiv.org); overlapping topics (base, model); pushes against this story (but).