Ai2 releases AstaBrief, an open 8B model for cited research reports

The short version

Ai2 released AstaBrief, a Qwen3-8B-based open-weight model that turns a research question and retrieved literature excerpts into a cited report. It powers Asta Fast mode; Ai2 also publishes training data and an example PDF workflow.

What changed

The release makes model weights, training data, and a demonstration workflow available for this report-generation task.

What it means for you

Researchers can inspect and adapt the report-generation approach, while independently checking references and distinguishing generation speed from whole-pipeline speed.

The timing figures are publisher-reported results for the evaluated workflow. Claims about a nearly order-of-magnitude speed gain concern the generation component, not a tenfold speedup of the full pipeline.

Citations still need source checking. The historical development period limits comparisons with current frontier systems. Event precision follows the October 2 article date.

Fact check

VerifiedAstaBrief is an open-weight model based on Qwen3-8B that accepts a question and retrieved literature excerpts to produce a cited report.Evidence

VerifiedIt powers Asta Fast mode alongside Claude-powered Thinking mode; model weights, training data, and an example PDF workflow are released.Evidence

VerifiedThe reported full-pipeline averages are 51.1 seconds for Fast and 178.5 seconds for Thinking, approximately 3.5 times faster for Fast in that evaluation.Evidence

VerifiedMost development took place in 2025, and Ai2 explicitly says the model was not evaluated against current frontier models.Evidence