AI & ML Efficiency Breakthrough

Any-order autoregressive models can outperform diffusion-based classifiers while being 25x more efficient.

March 20, 2026

Original Paper

Revisiting Autoregressive Models for Generative Image Classification

Ilia Sudakov, Artem Babenko, Dmitry Baranchuk

arXiv · 2603.19122

The Takeaway

By removing the restrictive fixed token order in visual AR models, this paper unlocks their generative classification potential, beating diffusion-based methods on accuracy and efficiency. This provides a high-performance alternative to diffusion models for robust image classification and generative tasks.

From the abstract

Class-conditional generative models have emerged as accurate and robust classifiers, with diffusion models demonstrating clear advantages over other visual generative paradigms, including autoregressive (AR) models. In this work, we revisit visual AR-based generative classifiers and identify an important limitation of prior approaches: their reliance on a fixed token order, which imposes a restrictive inductive bias for image understanding. We observe that single-order predictions rely more on p