Diffusion Language Models

How this thread developed

1. Sept 2026 · University of Texas at Austin

   Hybrid backbones can efficiently adapt into diffusion language models

   This paper adapts pretrained autoregressive models into diffusion LMs across scales, directly advancing the thread on diffusion-based generation adaptation.

2. 4 further papers

   Oct 2026 · Seoul National University, Harvard University

   Diffusion models beat autoregressive LLMs on globally constrained problems

   Shows masked diffusion language models outperform autoregressive LLMs on globally constrained tasks using a fixed checkpoint and mean confidence guidance.

3. Oct 2026 · NVIDIA, Cornell University

   Semi-autoregressive continuous diffusion models match discrete baselines and enable KV caching.

   Proposes a semi-autoregressive continuous diffusion framework for language that enables KV caching.

4. Oct 2026 · University of Illinois Urbana-Champaign, Amazon.com, Inc.

   Coupling continuous latent with token feedback improves diffusion language models

   Proposes a hierarchical hybrid of continuous and discrete diffusion for language, improving coherence over pure continuous approaches.

5. 1 further paper

   Oct 2026 · NVIDIA, Cornell University

   Continuous diffusion language models match discrete peers while enabling smoother control

   introduces a continuous diffusion LM that matches autoregressive models on math and coding

6. Oct 2026

   Parallel diffusion decoding rivals autoregressive visual grounding at 4B scale

   Applies parallel diffusion decoding to visual grounding, demonstrating blockwise non-autoregressive generation at scale.

7. 1 further paper

   Oct 2026 · The Hong Kong University of Science and Technology

   Repelling crossing trajectories makes flow matching smoother and easier to sample

   Presents CoFlow, a flow-matching method that repels crossing trajectories during training for smoother and easier sampling

7 of 13 papers shown