Open-sourcing AstaBrief, the fast report-generation model in Asta

Summary

AstaBrief is an open-weights 8B-parameter model developed by Ai2 for generating cited scientific reports from a research question and literature excerpts, designed for evidence-grounded outputs. It was trained from Qwen3-8B using supervised fine-tuning on 47K filtered real user queries and direct preference optimization, focusing on data quality. Unlike Asta’s Thinking mode, which uses a multi-step pipeline, AstaBrief writes the full report in a single pass to reduce generation time. In Asta’s Fast mode, it averages 51.1 seconds per report versus 178.5 seconds for Thinking mode, roughly 3.5× faster, while competitive on answer and citation quality. The model and training data are open-sourced on Hugging Face, enabling institutions to deploy behind their own firewall and supporting future open scientific models.

Takeaways

1. 1

   AstaBrief generates reports in a single pass, achieving roughly 3.5× faster generation than the multi-step Thinking mode (51.1 seconds vs 178.5 seconds average).

2. 2

   The model was trained with a focus on data quality over complex reinforcement learning, using SFT and DPO on real user queries and filtered training data.

3. 3

   Open-sourcing allows institutions to deploy AstaBrief on their own infrastructure behind a firewall, addressing privacy concerns for sensitive research.

FAQ