# Diffusion Language Model

> A language model that generates entire blocks of tokens by iterative denoising rather than one token at a time.

- Category: Models & Training
- Stage: Emerging
- Also called: dLLM, Text diffusion model
- Canonical: https://www.joelcsaji.com/ai-dictionary/diffusion-language-model

## Plain English

The text condenses out of noise a chunk at a time instead of being typed out word by word.

## Example

An in-editor completion built on a dLLM fills a 50-line function almost instantly, so the UI needs a reveal animation rather than a typing effect.

## Why it matters

The first serious architectural break from autoregressive generation in commercial LLMs - Gemini Diffusion samples at ~1,479 tokens per second, and streaming-style UIs stop making sense.

## Related

- [Latency](https://www.joelcsaji.com/ai-dictionary/latency)
- [Streaming](https://www.joelcsaji.com/ai-dictionary/streaming)
- [Frontier Model](https://www.joelcsaji.com/ai-dictionary/frontier-model)

## Sources

- [Google DeepMind: Gemini Diffusion](https://deepmind.google/models/gemini-diffusion/)