🔍 Read the full analysis: AstaBrief And The Move To Open-Source AI 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 designed to turn a research question and retrieved literature excerpts into a cited report. Ai2 reports an average generation time of 51.1 seconds in Asta’s Fast mode, versus 178.5 seconds in its Claude-powered Thinking mode, but has not rerun its full evaluation against current frontier models.
Ai2 has open-sourced AstaBrief 8B, a model designed to generate cited scientific reports from a research question and retrieved literature excerpts, as detailed in the original analysis. The release includes the model weights, training data and an example workflow; Ai2 says the model is also available as Fast mode in its Asta research platform, where its reported average generation time is shorter than the platform’s Claude-powered Thinking mode.
Ai2 says AstaBrief is built on Qwen3-8B and adapted for long-form scientific synthesis. Its pipeline takes a user’s question and relevant retrieved snippets and generates a report in one pass. According to Ai2, this skips snippet summarization and clustering steps used in Thinking mode, as well as writing the report section by section.
Across Asta’s full pipeline, Ai2 reports an average of 51.1 seconds per report in Fast mode, compared with 178.5 seconds in Thinking mode. Based on those figures, Fast mode took about 3.5 times less time. The figures describe generation time; they do not, on their own, establish that reports from the two modes have equivalent accuracy, coverage or citation quality.
The company says it trained AstaBrief with supervised fine-tuning and direct preference optimization, rather than reinforcement learning. Ai2 describes its approach as creating and filtering examples that demonstrate the report-writing behavior it wanted, including attention to citation grounding. The release also provides an example workflow for generating reports from researchers’ own PDFs.
Local Access to Scientific Reports
The release gives research groups a model they can download, inspect and adapt, rather than relying only on a hosted report-generation service. Ai2 says institutions could run open weights on their own infrastructure, which may suit work involving unpublished research or sensitive questions. Whether a local setup offers the same performance as Asta’s hosted Fast mode is not established in the supplied information.
For researchers, the reported speed difference could make it easier to generate and revisit literature reports as working documents. But speed is only one measure of usefulness. A report must also represent the evidence accurately, preserve limitations in the underlying studies and provide citations that readers can check. The open release makes independent testing possible; it does not itself show that those standards are met.
AI scientific report generation software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
How Asta’s Two Modes Differ
Asta is Ai2’s platform for scientific work. Ai2 says users ask it to compare research across a literature base while applying constraints such as a particular method, population or setting. Its report feature now includes AstaBrief Fast mode alongside Thinking mode, which Ai2 describes as Claude-powered.
The modes use different generation processes, according to the company. Fast mode generates a report directly from retrieved excerpts, while Thinking mode uses additional processing and writes the report in stages. Ai2 says most of its model development and evaluation work was completed in 2025, using proprietary models that reflected the frontier at that time. It has not rerun the full evaluation against current frontier models.
Ai2 also places the release within a broader effort to adapt open models for scientific work, including activity with research communities through the NSF OMAI initiative. The company says it expects to share further findings from that work.
““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.
Report Quality Still Needs Testing
Ai2 has not reported a full comparison with current frontier models. The release material also does not provide enough detail to independently establish the company’s quality comparison between Fast and Thinking modes. The reported time averages are specific, but they do not answer whether one mode produces more accurate or complete reports.
Other open questions include how Ai2 measured report quality, how often citations directly support the statements they accompany, and how performance changes across scientific fields and query types. The available information does not specify the hardware and configuration behind the timing averages or confirm whether locally run models perform identically to the version in Asta. Independent assessments will be needed to evaluate citation reliability and whether reports preserve the limits of the research they summarize.
As an affiliate, we earn on qualifying purchases.
Independent Evaluations and Follow-Up
Researchers can examine the released weights and training data and adapt Ai2’s example workflow to their own PDFs. Evaluations by research groups could test the model across disciplines, evidence standards and local deployment configurations, including whether its citations are verifiable and its summaries remain faithful to the cited studies.
Ai2 says it expects to share more findings from its wider work on open models for science. A refreshed head-to-head evaluation against current frontier models has not been reported, so readers should treat the existing timing figures as Ai2’s results from its described setup rather than a current, independent comparison of report quality.
scientific literature analysis 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 create cited scientific reports from a research question and retrieved literature excerpts. Ai2 says the release includes training data and an example workflow.
How fast is AstaBrief compared with Asta’s Thinking mode?
Ai2 reports averages of 51.1 seconds per report for Fast mode and 178.5 seconds for Thinking mode across Asta’s full pipeline. Those figures measure reported generation time, not the comparative accuracy or citation quality of the reports.
Can researchers run the model on their own infrastructure?
Ai2 says the open weights can let institutions run the model locally and has provided an example workflow for reports based on researchers’ own PDFs. The supplied information does not establish whether local installations perform identically to Asta’s Fast mode.
Has AstaBrief been compared with current leading models?
Ai2 has not rerun its full evaluation against current frontier models. The company says most of its development and evaluation work was completed in 2025, using comparisons with proprietary models available at that time.
Primary source: Hugging Face · via ThorstenMeyerAI.com
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
