Reproduce “DiffResearch: Accelerating Deep Research with Diffusion Language Models” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Alan-Barsag Gazzaev, Alexey Gavrilov, Mikhail Mozikov, Sergey Muravyov

DiffResearch tests diffusion generation in a deep-research pipeline, where agents plan, retrieve sources, and write cited reports. It compares diffusion and autoregressive backbones in the same setup, using a 100-task benchmark. On an H200, both diffusion configurations cut mean end-to-end time from 71.45 to about 11.9 seconds; measured RACE scores were similar, with no established quality gain.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.