🔍 Read the full analysis: How Hebbia Uses GPT-6 Astra To Enhance Financial Research on ThorstenMeyerAI.com
Get the latest gadgets delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
TL;DR
Hebbia says a prerelease version of GPT-6 Astra outperformed the best alternative it tested on source selection and presentation requirements in financial research tasks. The results come from company-run evaluations and do not establish accuracy, time savings or performance in live deals.
Hebbia says a prerelease build of OpenAI’s GPT-6 Astra outperformed the best alternative it tested on two financial research tasks: finding relevant supporting sources and producing presentations that follow a banker’s instructions. The company ran the evaluations in its Matrix platform in August and September 2026; the original analysis also notes that the results are company-reported and do not show how the model performs in live transactions or other workflows.
Hebbia tested models on financial research cases using what it described as the same cases and grading criteria for each model. The company said domain experts wrote or audited the criteria against the source documents. On source selection, GPT-6 Astra scored 13.6 percentage points above the best alternative tested at citing required types of sources. Hebbia also reported that 99.5% of cited excerpts were relevant to the question.
For presentation tasks, the model met 97.9% of applicable requirements, 14.4 percentage points above the best alternative on that measure, according to Hebbia. The tests covered selecting documents or Matrix cells to support an answer and generating native PowerPoint slides that followed a specified structure and guidelines. The company has not provided enough detail in the supplied account to assess the number of test cases, the full roster of competing models, scoring variability or how criteria were weighted.
Hebbia illustrated the workflow with a hypothetical banker reviewing roughly 2,000 files for a potential acquisition. Matrix could help examine customer concentration, contract terms and churn, combine company documents with peer filings and industry news, and draft a diligence summary and valuation presentation. The example is not described as a completed client deal. Hebbia says users can follow citations from answers and slide footnotes to source documents and passages.
Hebbia reports GPT-6 Astra gains in financial research
Company-run tests found improvements in source selection and presentation requirements. The results cover selected research cases, not live deals or other workflows.
What Hebbia tested
Tests in August and September 2026 used financial research cases and the same grading criteria for each model.
Find supporting evidence
GPT-6 Astra scored higher on citing required types of sources. Matrix links answers to documents for review.
Follow the banker’s brief
The model produced native PowerPoint slides measured against specified structure and guidelines.
From files to a deal summary
Hebbia’s hypothetical acquisition example shows how document research could feed a presentation.
Examine roughly 2,000 files for details such as customer concentration, contract terms and churn.
Bring in peer filings and industry news alongside the data room documents.
Create a diligence summary and valuation slides, then inspect citations and footnotes against their sources.
What the results establish
The reported scores describe performance under Hebbia’s evaluation.
- Domain experts wrote or audited the criteria.
- GPT-6 Astra ranked first among tested models on both measures.
- How results transfer to messy, changing live deal documents.
- Actual time saved, review error rates and independent replication.
Evidence and Review in Deal Work
The reported results address two connected tasks in document-heavy financial work: locating evidence across many files and turning research into a presentation that meets a brief. If the performance holds in practice, stronger source coverage could make it easier for a reviewer to trace a claim, while a draft that follows requirements could reduce some editing. Hebbia has not measured those potential effects in the material provided.
The distinction between the two measures matters. A presentation can meet formatting and structural requirements while still relying on a mistaken interpretation; a well-supported answer may still require substantial work before it is ready for a client. Professionals would need to check not just whether a citation exists, but whether the cited material supports the conclusion and uses the right reporting period. The company’s scores are evidence about its selected tests, not proof that generated work can replace review.
For finance teams weighing AI tools, the report offers a vendor’s account of model performance in a specific platform and task set. It does not establish that GPT-6 Astra is more accurate across financial research generally, nor that its use changes deal outcomes. Independent evaluations and results from actual deployments would help clarify whether the reported advantage translates into dependable work.
Products Worth Considering

Apple AirPods Pro

Samsung Odyssey OLED G8 34-Inch Curved Ultra WQHD Gaming Monitor

Bose QuietComfort Ultra Headphones (2nd Gen), Wireless Noise-Cancelling Over-Ear Headphones, Driftwood Sand
As an affiliate, we earn on qualifying purchases.
How Matrix Links Research to Sources
Matrix is Hebbia’s platform for bringing research together from documents and tying answers back to supporting material. In the company’s acquisition scenario, a user would work across files in a data room and could add external sources such as peer-company filings and industry news. The platform is presented as a way to assemble a diligence summary and presentation while retaining links to the underlying evidence.
That traceability is central to the workflow Hebbia describes: a user can open a citation to inspect a document and highlighted passage, and slide footnotes can link to source material. The stated purpose is to let professionals check answers before relying on them. The example involving an industrial-software company is hypothetical, and Hebbia did not report that the scenario was a live transaction or a measured customer result.
“GPT-6 Astra scored 13.6 percentage points above the best alternative tested on citing required types of sources.”
— Hebbia
Limits of the Company Tests
The results are from Hebbia’s own evaluation of a prerelease model build. The supplied account does not give the case count, complete model comparison set, variability across runs or weighting of the grading criteria. Without those details or an independent replication, it is not possible to judge how broadly the scores apply.
Hebbia also did not quantify time saved, report how often reviewers found errors, or describe measured effects on client decisions. Performance in a live deal may differ when documents are incomplete, inconsistent or updated during the process. The company’s expectations about faster review or less overnight work should not be read as reported deployment outcomes.
It is also unclear when GPT-6 Astra will be available for this workflow, what access conditions would apply, and whether the reported test version matches any future release. Hebbia has described a possible role for agents that monitor information and surface signals to finance professionals, but has supplied no release timeline or evaluation results for that proposed capability.
Broader Tests and Deployment Evidence
The next useful evidence would include fuller evaluation details, independent tests and results from finance professionals using the system on actual research work. Those reports could show how often generated claims withstand review, whether citations make checking more effective, and whether presentation drafting reduces work without introducing new errors.
Hebbia has also pointed to a future possibility in which agents monitor information and flag developments for professionals to assess. The company has not stated a launch date or provided performance data for that idea. Until more evidence is available, the reported figures should be understood as results from a company-run test rather than a measured account of live deal performance.
Key Questions
What did Hebbia test?
Hebbia tested a prerelease GPT-6 Astra build in Matrix on financial research tasks involving selection of supporting sources and creation of presentations that followed specified requirements.
What results did Hebbia report?
The company said GPT-6 Astra scored 13.6 percentage points higher than the best alternative tested on required source types and met 97.9% of applicable presentation requirements, 14.4 percentage points above that alternative.
Do the results show the model is reliable for live deals?
No. The figures come from Hebbia-run tests using selected cases and criteria. The supplied report does not establish performance in live transactions, where documents and research conditions may differ.
Did Hebbia report time savings or fewer errors?
No quantified time savings or error rates were provided. The company described possible workflow benefits, but did not report measured effects from client deployments in the supplied material.
What is Hebbia proposing for future development?
Hebbia has described the possibility of agents that monitor information and flag developments for finance professionals. It has not provided a release date or evaluation results for that proposed role.
Primary source: OpenAI · via ThorstenMeyerAI.com
