Research · Hacker News ·

Diffusion LLMs mature: open guides from Google, NVIDIA, Inception

A comprehensive research guide traces how diffusion models — the engine behind AI image generation — are being adapted for text, with production models now from Google, NVIDIA, and Inception Labs.

Based on reporting by Hacker News — analysis by dalili

Diffusion models conquered image generation. Now they're coming for text. A new comprehensive guide from the Kuleshov Research Group lays out the full technical stack behind diffusion language models — the alternative to autoregressive (left-to-right) generation that's gaining serious momentum in 2026.

Unlike standard LLMs that produce text one token at a time, diffusion language models generate the entire sequence at once, starting from random noise and iteratively refining it. This unlocks key advantages: built-in error correction (mistakes can be revised mid-generation), bidirectional context (every position attends to the full sequence), and potentially 5-10x faster inference via parallel generation.

The guide traces the evolution from simple masked diffusion (essentially a generative BERT) through modern production systems. Google's Gemma Diffusion, NVIDIA's Nemotron Diffusion, and Inception Labs' Mercury 2 all ship diffusion-based LLMs today. Mercury 2 achieves over 1,000 tokens/second on standard GPUs — speeds previously possible only with specialized inference chips.

The authors argue diffusion may unlock the next scaling leap: just as transformers unlocked pre-training scaling by parallelizing training, diffusion could unlock inference-time scaling by parallelizing generation.

Key takeaways

  • Diffusion LLMs generate entire sequences at once, enabling error correction
  • Mercury 2 hits 1,000+ tokens/sec on standard GPUs via parallel generation
  • Google, NVIDIA, and Inception Labs all ship production diffusion LLMs
  • Authors argue diffusion could parallelize inference like transformers parallelized training

Why it matters

Diffusion LLMs represent the most credible architectural challenge to autoregressive dominance in years. With production models from Google, NVIDIA, and a commercial startup all shipping, and speeds that rival specialized hardware, this isn't just academic — it's a potential paradigm shift for how we run AI at scale.