ByteDance releases open Axion 1.0 agent model trained in three stages
, updated First seen Updated Emerging, tapering
Models
Reported by letsdatascience and The Decoder2 sources ↓
ByteDance and researchers at three US universities released Axion 1.0, an open agent foundation model. The accompanying paper argues that three training stages can be enough to develop agents.
Key points
- Axion 1.0 is an open agent foundation model
- The model was trained in three stages
- Researchers from Duke, the University of Maryland, College Park, and the University at Buffalo collaborated with ByteDance
Sources
Details
ByteDance and researchers at Duke University, the University of Maryland, College Park, and the University at Buffalo released Axion 1.0 on November 21, 2026. The model is described as an open agent foundation model.
The accompanying paper argues that three training stages are sufficient for developing AI agents. It says increasingly complex training pipelines make it difficult to determine which steps help and can make work expensive to reproduce.
Written from coverage as of
Images
- The Decoder