🔍 Read the full analysis: Inside AstaBrief: Asta’s Model For Faster Report Generation on ThorstenMeyerAI.com
Get monitors, keyboards and dev gear 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 released AstaBrief 8B, an open-weights model for generating cited scientific reports from research questions and retrieved literature excerpts. Ai2 reports an average generation time of 51.1 seconds in Fast mode, versus 178.5 seconds for its Claude-powered Thinking mode, but has not rerun its full comparison against current frontier models.
Ai2 has released AstaBrief 8B, an open-weights model designed to generate cited scientific reports from a research question and retrieved literature excerpts. The model is also available as Fast mode in Ai2’s Asta platform, where the company reports an average of 51.1 seconds per report, compared with 178.5 seconds for its Claude-powered Thinking mode.
Ai2 says the release includes the model weights, training data and an example workflow that researchers can adapt to generate reports from their own PDFs. The model is based on Qwen3-8B and was adapted for long-form scientific synthesis. Its intended input is a query plus relevant retrieved excerpts; it then generates a report in one pass.
That pipeline differs from Asta’s Thinking mode, according to Ai2. The company says Thinking mode uses Claude and includes steps for summarizing and clustering snippets, followed by section-by-section report writing. AstaBrief Fast mode bypasses those stages. Using the reported averages, Fast mode takes about 3.5 times less time than Thinking mode; the figures describe Ai2’s full Asta pipeline, not a general speed comparison across models or settings.
Ai2 says it trained AstaBrief with supervised fine-tuning and direct preference optimization, using selected examples intended to teach report-writing behavior and citation grounding. The team considered reinforcement learning but used the other methods. The announcement describes a focus on filtering examples and preference data; it does not provide enough detail to independently establish that the model matches Thinking mode in report quality.
Faster Reports, Local Control
The release gives research groups an inspectable model and workflow for producing literature-based reports, rather than access only through a hosted feature. Institutions may be able to run open weights on their own infrastructure, which could be useful when questions or documents involve unpublished work or sensitive research. Ai2’s example workflow also offers a starting point for adapting report generation to a group’s own PDFs.
The reported time difference could make it easier to generate and revisit reports as working research aids. But speed is not a measure of accuracy. A short turnaround does not show whether a report covers relevant studies, represents their limitations correctly, or links claims to supporting citations. Those checks matter if researchers use generated reports to guide further reading or compare findings.
Open weights and training materials allow outside researchers to examine and test the system, but they do not by themselves establish that local deployments reproduce the service’s reported timing or quality. The practical value will depend on how reliably the model handles different disciplines and evidence standards, and on whether its references can be checked against the source literature.
scientific report generation software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
How Asta’s Report Modes Differ
Asta is Ai2’s platform for scientific work. Ai2 says users ask it to compare approaches across research literature while applying constraints such as a particular method, population or setting. Its report-generation feature now offers AstaBrief Fast mode alongside the Claude-powered Thinking mode.
The two modes use different generation processes, as described by Ai2: Fast mode produces a report in one pass from a query and retrieved snippets, while Thinking mode adds intermediate synthesis and writes the report section by section. Ai2 frames the shorter process as a way to reduce generation time while retaining report quality, but the provided material does not establish that quality comparison in enough detail for independent assessment.
Ai2 says most of the training and evaluation work was completed in 2025, using proprietary models available during that period. The company has not rerun its full evaluation against current frontier models. It has also described AstaBrief as part of a broader effort 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
As an affiliate, we earn on qualifying purchases.
Quality Checks Still Needed
Ai2 has not published a full comparison against current frontier models, and its supplied material does not spell out enough about report-quality measurement to assess the claimed quality trade-off. The timing averages are reported by Ai2; the announcement does not specify the hardware and configuration behind them.
Other open questions include how often citations directly support the statements they accompany, how performance varies across disciplines and query types, and whether a locally run model performs like Fast mode in Asta. The release makes the weights and training data available, but independent evaluation is still needed to determine accuracy, coverage and citation reliability across uses.
literature review automation software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Independent Testing and Updates
Researchers can inspect the released weights and training data and adapt Ai2’s example workflow to reports based on their own PDFs. Testing by research groups could help establish how the model performs across subjects, evidence standards and local deployment configurations.
Ai2 says it expects to share more findings from its broader research on adapting open models to scientific work. A new full comparison with current frontier models has not been reported, so the next useful evidence would include updated head-to-head evaluations, transparent quality measures and checks of whether citations support the report’s claims.
As an affiliate, we earn on qualifying purchases.
Key Questions
What is AstaBrief 8B?
AstaBrief 8B is Ai2’s open-weights model for generating scientific reports from a research question and retrieved literature excerpts. Ai2 says it is based on Qwen3-8B and adapted for long-form synthesis.
How fast is AstaBrief Fast mode?
Ai2 reports an average of 51.1 seconds per report for Fast mode across Asta’s full pipeline. The company reports 178.5 seconds for its Claude-powered Thinking mode; these are Ai2’s figures, not an independently reported benchmark.
Does the release show that AstaBrief produces reports as well as Thinking mode?
No. Ai2 presents Fast mode as a way to reduce generation time while maintaining quality, but the supplied information does not provide enough detail to independently establish that comparison. Speed alone does not demonstrate report accuracy or citation quality.
Can researchers run AstaBrief on their own documents?
Ai2 released an example workflow intended for generating reports from researchers’ own PDFs, along with the model weights and training data. The announcement does not establish whether local deployments match Asta’s service in speed or quality.
Has Ai2 compared the model with current frontier systems?
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.
Primary source: Hugging Face · via ThorstenMeyerAI.com
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
