DiffResearch: Accelerating Deep Research with Diffusion Language Models
Summary
DiffResearch 将自回归骨干替换为扩散语言模型（dLLM）用于多智能体深度研究，在 Deep Research Bench 上实现约 6 倍端到端加速并提升报告质量，还提出 dFast 提前从去噪状态提取子查询，以 Apache-2.0 开源。
View Cached Full Text
Cached at: 10/22/26, 07:36 PM
Paper page - DiffResearch: Accelerating Deep Research with Diffusion Language Models
Source: https://huggingface.co/papers/2610.28909
Abstract
DiffResearch replaces autoregressive backbones with diffusion language models in multi-agent deep research, achieving large latency reductions and improved report quality via iterative planning and early subquery extraction.
Deep research systems iteratively plan, gather evidence, synthesize, and revise, making end-to-end latency a key bottleneck for interactive use. Almost all production systems are built on autoregressive (AR) language models and inherit left-to-right decoding in planning and long-form writing. We present DiffResearch, an openmulti-agent deep researchframework that places adiffusion language model(dLLM) at the core of every stage, and measure how adiffusion backbonechanges latency and report quality. Holding the agentic scaffold, retrieval code, prompts, and evaluation fixed, we swap the backbone between an open diffusion model and a comparable AR model. Across 100 Deep Research Bench tasks, thediffusion backboneis roughly 6 times faster end-to-end, driven by order-of-magnitude reductions in planning and writing. We further introducedFast, a decomposition mode that reads intermediatedenoising statesand extracts subqueries before generation finishes, providing a 1.99 times median speedup for decomposition in isolation with no loss of report quality. On the benchmark’s RACE rubric,dFastobtains the best overall score among the three configurations. Finally, in a public system-level leaderboard snapshot, an instance built on a stronger diffusion model ranks above several commercial systems. We release DiffResearch under an Apache-2.0 license.
View arXiv pageView PDFGitHub41Add to collection
Get this paper in your agent:
hf papers read 2610\.28909
Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash
Models citing this paper0
No model linking this paper
Cite arxiv.org/abs/2610.28909 in a model README.md to link it from this page.
Datasets citing this paper0
No dataset linking this paper
Cite arxiv.org/abs/2610.28909 in a dataset README.md to link it from this page.
Spaces citing this paper0
No Space linking this paper
Cite arxiv.org/abs/2610.28909 in a Space README.md to link it from this page.
Collections including this paper0
No Collection including this paper
Add this paper to acollectionto link it from this page.
Similar Articles
On-Policy Residual Learning for Diffusion Language Models
该论文提出ORL（On-policy Residual Learning）框架，通过熵收缩目标与异步残差捕获等系统级协同设计，使扩散语言模型（DLM）在相近质量下将吞吐量提升至约6.9×，在单请求解码速度上超越自回归模型的推测解码方案。
$R^2$-dLLM: Accelerating Diffusion Large Language Models via Spatio-Temporal Redundancy Reduction
R²-dLLM introduces spatio-temporal redundancy reduction techniques that cut diffusion LLM decoding steps by up to 75% while preserving generation quality, addressing a key deployment bottleneck.
Diff Diffusion: Predicting Edit Paths with Diffusion Models for Efficient Code Editing
DiffDiffusion提出一种受扩散模型启发的LLM框架，通过预测编辑路径而非重新生成整个代码段来实现高效代码编辑，在公共基准上比自回归基线快2.78倍，并在生产环境中将P50模型延迟降低39.0%。
Dynamic-dLLM: Dynamic Cache-Budget and Adaptive Parallel Decoding for Training-Free Acceleration of Diffusion LLM
This paper proposes Dynamic-dLLM, a training-free framework that accelerates diffusion large language models by dynamically allocating cache-update budgets and calibrating decoding thresholds, achieving over 3x speedup on models like LLaDA and Dream while maintaining performance.
@_akhaliq: Unlocking Lossless Speedups in LLMs via Discrete Diffusion paper: https://huggingface.co/papers/2609.04010…
This paper proposes a method using discrete diffusion to unlock lossless speedups in large language models, aiming to improve efficiency without compromising performance.