English

Evo: Autoregressive-Diffusion Large Language Models with Evolving Balance

Machine Learning 2026-03-10 v1 Artificial Intelligence

Abstract

We introduce \textbf{Evo}, a duality latent trajectory model that bridges autoregressive (AR) and diffusion-based language generation within a continuous evolutionary generative framework. Rather than treating AR decoding and diffusion generation as separate paradigms, Evo reconceptualizes text generation as a latent flow: each token is associated with a vector-valued embedding that evolves over a progression variable ti[0,1]t_i \in [0, 1], indicating its semantic maturity. Low tit_i values correspond to confident AR-like refinement, while high values invoke diffusion-style planning, allowing the model to adaptively balance AR and diffusion based on uncertainty. Theoretically, we show that both AR and diffusion models emerge as discretizations of a shared probability flow, and we derive Evo's training objective from a unified variational ELBO. The model is implemented as a time-conditioned Transformer governed by a shared vector field, trained end-to-end to jointly infer latent codes and their progression times. During decoding, Evo performs efficient, semantics-aware refinement, achieving high-quality outputs without sacrificing speed. Empirically, Evo 8B achieves state-of-the-art or highly competitive results on 15 diverse benchmarks, including reasoning (GSM8K, ARC-C), code generation (HumanEval, MBPP), and general language understanding, while maintaining fast inference speed. Our results demonstrate that Evo delivers a new paradigm for LLM design with strong generation quality, robust symbolic reasoning, and decoding efficiency.

Keywords

Cite

@article{arxiv.2603.06617,
  title  = {Evo: Autoregressive-Diffusion Large Language Models with Evolving Balance},
  author = {Junde Wu and Minhao Hu and Jiayuan Zhu and Yuyuan Liu and Tianyi Zhang and Kang Li and Jingkun Chen and Jiazhen Pan and Min Xu and Yueming Jin},
  journal= {arXiv preprint arXiv:2603.06617},
  year   = {2026}
}
R2 v1 2026-07-01T11:07:32.847Z