🔍 Read the full analysis: How AstaBrief Powers Fast Report Generation At Asta 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 that Asta’s Fast mode averaged 51.1 seconds per report, versus 178.5 seconds for 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 turn a research question and retrieved scientific literature excerpts into a cited report, as detailed in the original analysis. The model is also available in Ai2’s Asta platform as Fast mode, where Ai2 reports an average generation time of 51.1 seconds, compared with 178.5 seconds for Asta’s Claude-powered Thinking mode.
Ai2 says AstaBrief is built on Qwen3-8B and adapted for long-form scientific synthesis. The release includes the model weights, training data and an example workflow that researchers can modify to generate reports from their own PDFs. Open weights give institutions the option to examine and adapt the model, including for use on their own infrastructure.
In Asta, Fast mode takes a user query and relevant retrieved excerpts and generates a report in one pass, according to Ai2. The company says this bypasses stages used by Thinking mode, including summarizing and clustering excerpts and writing the report section by section. Ai2 attributes the lower reported generation time to this more direct pipeline.
Ai2 says AstaBrief was trained with supervised fine-tuning and direct preference optimization, rather than reinforcement learning. The team focused on creating and filtering examples that model the desired report-writing behavior, with particular attention to citation grounding. The material supplied by Ai2 describes the approach but does not give enough detail to independently establish that Fast mode’s reports match Thinking mode in quality.
Faster Reports, Local Research Options
The release gives research groups an inspectable model and a sample workflow for producing literature-based reports, rather than access only through a hosted service. For institutions handling unpublished work or sensitive research questions, local deployment could offer more control over where queries and documents are processed. Whether a local installation provides the same performance as Asta’s hosted Fast mode is not specified.
The reported time difference could make it easier to generate and revisit reports as working research documents. Ai2’s figures show an average of 51.1 seconds in Fast mode and 178.5 seconds in Thinking mode across Asta’s full pipeline, or about 3.5 times faster by those averages. That is a speed comparison, not evidence that the faster reports are equally accurate or complete. Scientific users still need to check whether each citation supports the claim attached to it and whether the report preserves the limits of the underlying studies.
scientific report writing software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
How Asta’s Two Modes Differ
Ai2 describes Asta as an agentic platform for scientific work. The company 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 Thinking mode, which Ai2 describes as Claude-powered.
The modes use different report-generation pipelines. According to Ai2, Thinking mode summarizes and clusters retrieved excerpts before drafting a report section by section. Fast mode instead sends the query and retrieved excerpts into a single report-generation step. Ai2 says its model training used real research queries and preference data, with filtering intended to encourage citation-focused writing.
Ai2 says most of the training and evaluation work was completed in 2025, using proprietary models that reflected the frontier at that time. The company has not rerun the full evaluation against current frontier models, so the announcement’s speed figures should not be read as a current comparison of report quality with those systems.
““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’s announcement reports generation times but does not provide enough information here to independently verify the quality comparison between Fast and Thinking modes. The material does not specify how report quality was measured, how often citations were checked and found to support their associated claims, or how results vary across disciplines and query types.
There is also no reported full evaluation against today’s frontier models. The hardware and configuration behind the timing averages are not detailed in the supplied material, and it is unclear whether locally run versions of AstaBrief perform identically to the version in Asta. The speed results are Ai2-reported averages, not a guarantee that every query will produce a report in the same time.
These gaps matter because a report can be fast while still omitting relevant studies, misrepresenting findings or citing sources that do not substantiate its statements. Independent evaluation would be needed to establish how reliably AstaBrief handles those risks across fields and deployment settings.
As an affiliate, we earn on qualifying purchases.
Independent Tests and Further Findings
Researchers can examine the released weights and training data and adapt Ai2’s example workflow to generate reports from their own PDFs. Testing across disciplines, query types and local computing setups could help establish where the model performs well and where additional review is needed.
Ai2 says the work is part of a broader effort to adapt open models for scientific needs, including work with scientific communities through the NSF OMAI initiative, and says it expects to share more findings. Ai2 has not announced a date for a new evaluation against current frontier models. Until such comparisons and independent citation checks are available, the reported timing advantage is the clearest quantified result in the material.
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 built on Qwen3-8B and adapted for long-form synthesis.
How fast is Asta’s Fast mode?
Ai2 reports an average of 51.1 seconds per report across Asta’s full pipeline. Its Claude-powered Thinking mode averaged 178.5 seconds in the same reported comparison; these figures do not establish a difference in report quality.
Can researchers run AstaBrief locally?
Ai2 has released the model weights and an example workflow researchers can adapt, including for reports based on their own PDFs. The supplied material does not detail required hardware or confirm that local performance matches Asta’s hosted Fast mode.
Has Ai2 compared AstaBrief with current frontier models?
Not in a newly rerun full evaluation, according to Ai2. The company says most of the evaluation work was completed in 2025 and has not reported a full comparison against current frontier models.
Does the reported speed show that AstaBrief’s reports are accurate?
No. Generation time measures speed, not whether a report is complete, accurately represents studies or uses citations that support its claims. The supplied material leaves those quality measures and independent results unclear.
Primary source: Hugging Face · via ThorstenMeyerAI.com
Fall Picks
fall essentials
As an affiliate, we earn on qualifying purchases.
