TL;DR
Get tech for your team 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
Ai2 has open-sourced AstaBrief 8B, an open-weights model designed to turn a research question and retrieved literature excerpts into a cited report. Ai2 reports that Asta’s Fast mode averaged 51.1 seconds per report, compared with 178.5 seconds for its Claude-powered Thinking mode; it has not rerun its full evaluation against current frontier models.
Ai2 has open-sourced AstaBrief 8B, a model that generates cited scientific reports from a research question and retrieved literature excerpts. The model is also available in Asta’s report-generation feature as Fast mode; Ai2 says the release includes training data and an example workflow that researchers can adapt for local report generation.
Ai2 says AstaBrief is built on Qwen3-8B and was adapted for long-form scientific synthesis. Its training used supervised fine-tuning and direct preference optimization, rather than the reinforcement-learning approach the team considered. Ai2 says it focused on creating and filtering examples that demonstrated the desired report-writing behavior, including attention to citation grounding.
The model’s generation pipeline takes a user query and relevant retrieved snippets and produces a report in one pass. Ai2 says this design bypasses snippet summarization and clustering stages used in Asta’s Claude-powered Thinking mode, as well as section-by-section report writing. The company presents the approach as a way to reduce generation time while maintaining report quality, but its supplied material does not provide enough detail here to independently establish that quality comparison.
Across Asta’s full pipeline, Ai2 reports an average of 51.1 seconds per report in Fast mode, compared with 178.5 seconds for Thinking mode. That is about 3.5 times faster based on the figures given. Ai2 also says open weights could let institutions run the model on their own infrastructure, a feature relevant to research involving sensitive or unpublished questions.
Faster Reports With Local Deployment
The release gives researchers and institutions access to a model designed for cited literature synthesis that they can download and adapt. Local deployment may be useful when questions or documents involve unpublished research or other sensitive material. Ai2 also provides an example workflow for generating reports from researchers’ own PDFs, extending the release beyond the model weights alone.
The reported speed difference could make generated reports more practical as working research artifacts that users revisit and refine. Yet speed does not establish that a report is accurate, complete, or faithful to the underlying studies. Scientific synthesis depends on preserving the limits of the evidence and making citations verifiable, so readers will need to judge the model’s outputs on those measures as well as response time.
scientific report generation software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Asta’s Two Report Modes
Asta is Ai2’s agentic platform for scientific work. The company says users often ask it to compare approaches across research literature while applying constraints such as a particular method, population, or setting. Its Generate a report feature now offers AstaBrief Fast mode alongside Thinking mode, which Ai2 describes as Claude-powered.
Ai2 says the model was trained using real research queries, citation-focused filtering, and preference data. The training and evaluation approach described in the announcement was developed against proprietary models that reflected the frontier at the time. Most of that work was completed in 2025, Ai2 says, and the company has not rerun the full evaluation against today’s frontier models.
““We wanted to help scientists generate cited reports faster, with a model they could download and run themselves.””
— Ai2
research literature citation tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Quality Comparisons Need Updating
Ai2 has not reported a full evaluation against current frontier models. Its comparisons use proprietary models from the period when most of the work was done, which Ai2 says was 2025. The announcement’s timing figures are specific, but readers should not treat them as a current head-to-head quality comparison.
The supplied material also leaves questions about how report quality was measured, how often citations accurately support the statements they accompany, and how performance varies across fields and query types. It does not establish whether locally run versions perform identically to the service’s Fast mode or detail the hardware and configuration behind the reported timing averages.
As an affiliate, we earn on qualifying purchases.
Testing the Open Release
Researchers can examine the released model weights and training data, then adapt Ai2’s example workflow for reports based on their own PDFs. Further evaluation by research groups could show how well AstaBrief handles different disciplines, evidence standards, and local deployment settings.
Ai2 says the announcement describes work on a broader effort to adapt open models to scientific needs, including work with scientific communities through the NSF OMAI initiative. The company says it expects to share more findings from that research. A current comparison with frontier models, however, has not yet been reported.
As an affiliate, we earn on qualifying purchases.
Where I land
I see AstaBrief as a useful open release because it pairs a model with training data and a workflow that researchers can inspect and adapt. The reported 51.1-second average suggests the one-pass design may make cited reports quicker to generate within Asta, while open weights create an option for institutions that want to run the model locally.
The strongest counterargument is that speed and openness do not establish trustworthy scientific synthesis. A fast report can still misrepresent a study, omit relevant evidence, or attach citations that do not support its claims. I would raise my assessment if independent evaluations across fields showed that AstaBrief consistently preserves the limits of the literature and produces verifiable citations, including in local deployments. A current comparison with frontier models would also help establish how its quality and speed trade off today.
Key Questions
What is AstaBrief 8B?
AstaBrief 8B is Ai2’s open-weights model for generating cited scientific reports from a research question and retrieved literature excerpts. Ai2 says it is based on Qwen3-8B.
How fast is AstaBrief in Asta?
Ai2 reports that Fast mode averaged 51.1 seconds per report across Asta’s full pipeline. Its Claude-powered Thinking mode averaged 178.5 seconds in the comparison Ai2 provided.
What has Ai2 released?
Ai2 says it is releasing the model weights and training data, as well as an example workflow researchers can adapt to generate reports from their own PDFs.
Has AstaBrief been compared with today’s frontier models?
No full evaluation against current frontier models is reported. Ai2 says most of the training and evaluation described was completed in 2025 and that the comparison models reflected the frontier at that time.
Source: Hugging Face
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
