# Plan-and-Patch boosts plan repair with diffusion language models

> Researchers show a diffusion‑based planner repairs plans twice as often as a traditional autoregressive model and cuts generation time by up to half.

Oossa · 2026-10-09 · https://oossa.com/en/plan-and-patch-boosts-plan-repair-with-diffusion-language-models

The paper introduces Plan-and-Patch, a framework where a diffusion language model (dLLM) creates a structured plan and then fixes only the broken parts. In tests on the Natural Plan benchmark, the diffusion planner repaired plans 53.7% of the time, compared with 27.0% for an autoregressive planner. After training on agentic tasks like ALFWorld, both approaches generated plans at similar success rates, but the diffusion method was 39‑46% faster. The work suggests a way to make long‑horizon agents more efficient and adaptable.

## The facts

- Plan‑and‑Patch repair success: 53.7% (diffusion) vs 27.0% (AR)
- Mean plan‑generation latency reduced by 39‑46% after task‑specific training

## Why it matters

Faster, more reliable plan repair could let autonomous assistants finish complex tasks with fewer interruptions.

## Sources & references

1. [Plan-and-Patch: Diffusion Language Models for Agentic Planning](https://arxiv.org/abs/2610.10786) – arXiv, 2026-10-09

Last updated: 2026-10-09