Next-Concept Modeling: Fast, Parallel Coarse-to-Fine Text Generation
Hermann Kumbong*
Stanford University
Ishan S. Khare*
Stanford University
Iddah Mlauzi Stanford University
Andrew Woen Stanford University
Charles Ding Stanford University
E. Kelly Buchanan Stanford University
Simran Arora Stanford University
Preprint, 2026
Abstract
Next-Concept Modeling (NCM) is a language modeling framework that generates text from coarse concepts to fine details. Unlike autoregressive models that generate one token at a time, or diffusion models that repeatedly refine the entire sequence, NCM predicts increasingly detailed representations of text in parallel. Across unconditional generation, translation, summarization, and reasoning tasks, NCM achieves competitive or better quality than strong diffusion baselines while generating text 10–31x faster and using 4–5x fewer training tokens.