# Diffusion Language Models Change Speed Across Tasks and Languages

Authors: Mohsen Fayyaz, Ali Modarressi, Daniel Israel, Lucas Bandarkar, Nanyun Peng.

Measures how many forward passes diffusion language models spend per output token across tasks and languages. Relates this cost to output copying and novelty, and tests whether answer-decoding speed can improve voting over sampled math answers.

First public: 2026-10-19.
ArXiv initial submission: 2026-10-19.

[Original paper](https://arxiv.org/abs/2610.25835) · [PDF](https://arxiv.org/pdf/2610.25835) · [HTML](https://arxiv.org/html/2610.25835v1) · [Reading-list entry](https://latent-implicit-thinking.github.io/papers/2610.25835/)
[Preprint BibTeX](https://latent-implicit-thinking.github.io/papers/2610.25835/citation.bib) · [Preprint CSL-JSON](https://latent-implicit-thinking.github.io/papers/2610.25835/citation.csl.json) · [JSON record](https://latent-implicit-thinking.github.io/papers/2610.25835/record.json)

## Approaches

- [Alternative generation methods](https://latent-implicit-thinking.github.io/topics/alternative-reasoning-schedules/): Change how intermediate steps are generated, revised, or searched—for example, by refining several steps in parallel. Some produce written chains of thought.
- [Analysis and evaluation](https://latent-implicit-thinking.github.io/topics/implicit-reasoning-evaluation/): Measure how well these methods work, investigate what their internal computation does, and identify where they fail. This includes experiments, studies of model internals, and mathematical results.

Reading-list version 0.11.0; literature checked through 2026-10-26.