Ai2 has open-sourced AstaBrief 8B, a small language model designed for generating cited scientific reports from research questions and retrieved literature excerpts. The model is available alongside its training data to allow researchers to study, reproduce, and build upon the approach.

  • Built on Qwen3-8B and trained using supervised fine-tuning (SFT) and direct preference optimization (DPO), AstaBrief generates full reports in a single pass rather than section by section.
  • The training pipeline utilized tens of thousands of real research queries, resulting in 47K SFT examples and approximately 6K DPO pairs filtered for quality and agreement between judge models.
  • In the Asta platform, the Fast mode powered by AstaBrief averages 51.1 seconds per report compared to 178.5 seconds for the Claude-powered Thinking mode, representing a 3.5x speed improvement.
  • The release includes an example workflow that enables institutions to run local report generation from their own PDFs, addressing needs for sensitive or unpublished work.

Open weights allow institutions to deploy the model on their own infrastructure, ensuring data privacy and control over research artifacts.