🔍 Read the full analysis: AstaBrief: Open-Sourcing Asta’s Fast Report-Generation Model on ThorstenMeyerAI.com
Get the latest gadgets delivered free — and shop member deals
- Fast, free delivery on millions of items
- Access to Prime Big Deal Days deals on October 6–7
- Prime Video, Amazon Music and more included
TL;DR
Ai2 has open-sourced AstaBrief 8B, a model that generates cited scientific reports from a research question and retrieved literature excerpts. Ai2 reports an average of 51.1 seconds per report in Asta’s Fast mode, but says it has not rerun its full evaluation against current frontier models, and the release materials do not establish an independent quality comparison.
Ai2 has open-sourced AstaBrief 8B, a model designed to turn a research question and retrieved literature excerpts into a cited scientific report. The release includes model weights, training data and an example workflow, as described in the original analysis; Ai2 also reports that the model’s Fast mode averaged 51.1 seconds per report in Asta, compared with 178.5 seconds for its Claude-powered Thinking mode.
Ai2 says AstaBrief is built on Qwen3-8B and adapted for long-form scientific synthesis. Its generation pipeline takes a user’s question and relevant retrieved snippets, then produces a report in one pass. In contrast, Ai2 says Asta’s Thinking mode uses Claude and includes snippet summarization and clustering, followed by section-by-section report writing.
Ai2 describes its training as supervised fine-tuning and direct preference optimization. The team focused on creating and filtering examples that demonstrate the desired report-writing behavior, including attention to citation grounding, rather than using the reinforcement-learning approach it had considered. Ai2 says the training used real research queries and preference data.
The company’s reported timings cover Asta’s full pipeline: 51.1 seconds on average for Fast mode and 178.5 seconds for Thinking mode. Those figures make Fast about 3.5 times quicker based on the reported averages. They describe speed, however, and do not by themselves establish that the two modes produce reports of comparable accuracy or completeness.
Open Weights for Research Workflows
The release gives researchers access to a model they can inspect, download and adapt, rather than limiting use to Asta’s hosted feature. Ai2 says institutions can run the open-weights model on their own infrastructure, which may suit work involving unpublished research or sensitive questions. Its example workflow for reports based on researchers’ own PDFs provides a starting point for testing the system against local materials.
The speed figure may also matter for researchers who use generated reports as working documents and want to revise or revisit them. Still, speed and openness are not measures of scientific reliability. A useful report must represent the evidence fairly, preserve the limits of individual studies and cite sources that support the statements made. Those qualities require evaluation beyond generation time.
scientific report writing software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Asta’s Fast and Thinking Modes
Asta is Ai2’s platform for scientific work. Ai2 says users ask it to compare research approaches while applying constraints such as a particular method, population or setting. The platform’s report-generation feature now offers AstaBrief Fast mode alongside Thinking mode, which Ai2 describes as Claude-powered.
Ai2 says most of the model work and evaluation described in the release was completed in 2025, using proprietary models that reflected the frontier at that time. The company has not rerun its full evaluation against current frontier models. The release is also part of a broader effort Ai2 describes to adapt open models for scientific needs, including work with scientific communities through the NSF OMAI initiative.
““We wanted to help scientists generate cited reports faster, with a model they could download and run themselves.””
— Ai2
research literature management tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Report Quality Still Needs Testing
The announcement does not provide enough detail to independently establish how AstaBrief’s report quality compares with Thinking mode or other systems. Ai2 has not reported a full evaluation against today’s frontier models, and the supplied material leaves open how quality was measured, how often citations support the claims they accompany, and how results vary across disciplines and query types.
The reported timings also do not specify the hardware and configuration behind the averages. It is not clear whether locally run models perform identically to Asta’s Fast mode in the service. Researchers will need to test whether reports are accurate, sufficiently complete and faithful to the evidence, rather than relying on speed as a proxy for those qualities.
As an affiliate, we earn on qualifying purchases.
Independent Tests of the Release
Researchers can examine the released weights and training data and adapt Ai2’s example workflow to produce reports from their own PDFs. Evaluation by independent groups could help show how the model performs across fields, evidence standards and local deployment setups. Ai2 says it expects to share more findings from its broader research on adapting open models to scientific work.
A current comparison with frontier models has not yet been reported. Until such results are available, the announcement supports a specific conclusion about access and reported speed, but not a general conclusion that AstaBrief matches or exceeds newer systems in scientific report quality.
scientific citation management software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What is AstaBrief 8B?
AstaBrief 8B is an open-weights model from Ai2 designed to generate cited scientific reports from a research question and retrieved literature excerpts. Ai2 says it is based on Qwen3-8B.
What did Ai2 release?
Ai2 says the release includes the model weights, training data and an example workflow researchers can adapt, including for reports based on their own PDFs.
How fast is AstaBrief’s Fast mode?
Ai2 reports an average of 51.1 seconds per report across Asta’s full pipeline. It reports 178.5 seconds for the Claude-powered Thinking mode; the figures compare reported generation times, not independently established report quality.
Has Ai2 compared AstaBrief with current frontier models?
Not in a full updated evaluation, according to Ai2. The company says most of its work was completed in 2025 and that it has not rerun the full comparison against current frontier models.
Does the release prove the reports are reliable?
No. The release provides access to the model and reports timing results, but the supplied material does not independently establish citation accuracy, completeness or performance across fields. Those qualities remain subjects for further testing.
Primary source: Hugging Face · via ThorstenMeyerAI.com
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
