🔍 Read the full analysis: What GPT-6 Astra Brings To Financial Research At Hebbia 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 reports that a prerelease version of OpenAI’s GPT-6 Astra outperformed the best alternative it tested on source selection and presentation requirements in financial research cases. The tests were run by Hebbia in August and September 2026; they do not establish performance in live deals or across other workflows.
As detailed in the original analysis, Hebbia says a prerelease build of OpenAI’s GPT-6 Astra outperformed the best alternative it tested on two financial research tasks: selecting supporting sources and creating presentations that follow a user’s brief. The company ran the tests in Matrix, its document research platform, and says the results apply to its selected cases and scoring criteria—not necessarily to live deal work or other uses of the model.
In tests conducted in August and September 2026, Hebbia compared models using the same financial research cases and grading criteria. The company said domain experts wrote or audited the criteria against the underlying documents. On source selection, GPT-6 Astra scored 13.6 percentage points above the best alternative tested for citing required types of sources. Hebbia also reported that 99.5% of the model’s cited excerpts were relevant to the question.
For presentation tasks, GPT-6 Astra met 97.9% of applicable requirements, 14.4 percentage points above the best alternative on that measure, according to Hebbia. The tasks tested whether a model could select documents or Matrix cells to support an answer and produce native PowerPoint slides that followed a requested structure and guidelines, a workflow with some parallels to GPT-6 Astra’s video workflow.
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, incorporate peer filings and industry news, then produce a diligence summary and valuation presentation. The company says citations in answers and slide footnotes link back to source documents, allowing a reviewer to inspect the supporting passages and figures, much as GPT-6 Astra can trace software bugs. The example was not described as a completed client deal.
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.
What the Finance Test Results Show
The reported results address two connected tasks in financial research: locating evidence across a large document set and turning analysis into a presentation that meets a brief. If those capabilities hold up in practice, they could reduce some of the manual work involved in assembling diligence materials and checking where claims came from. Hebbia has not, however, reported measured time savings or evidence that the tests changed a deal decision.
Traceable citations matter because a polished presentation is not sufficient if its claims are unsupported or omit relevant evidence. A banker still needs to check whether a cited passage supports the conclusion, and whether a figure—such as customer concentration—uses the right reporting period. Hebbia’s numbers indicate performance under its own evaluation, not a basis for treating generated analysis as verified.
The results are most useful as a company account of how one financial research platform is evaluating AI for document-heavy work. They do not yet show whether the reported balance between source coverage and presentation quality persists with different datasets, incomplete files or changing documents. Independent testing and evidence from actual users would help establish how much practical value the results represent.
Products Worth Considering

Sonos Five Wireless HiFi Speaker, Black

Samsung 27-Inch Odyssey G5 QHD Gaming Monitor (G50F)

Apple AirPods Pro 3 Wireless Earbuds with Active Noise Cancellation
As an affiliate, we earn on qualifying purchases.
How Matrix Fits Deal Research
Hebbia’s Matrix platform brings together information from materials such as financial statements, contracts and management presentations, with answers linked to supporting documents. In the company’s acquisition scenario, a user could combine a data room with outside material such as peer-company filings and industry news, then use the platform to prepare a diligence summary and slides based on a firm template.
The workflow described by Hebbia pairs drafting with review: users can follow a citation to a document and highlighted passage, while presentation footnotes can link to source material. That design aims to make it easier for finance professionals to check the evidence behind an answer. The company’s reported tests focused on source selection and adherence to presentation requirements; they do not establish the accuracy of every conclusion in a full deal process.
Hebbia also described a possible future use for agents that monitor information and surface signals for finance professionals. That remains a forward-looking idea in the material provided, rather than a feature with reported deployment results or a stated release date.
“99.5% of cited excerpts were relevant to the question.”
— Hebbia
What the Tests Do Not Establish
Hebbia has not provided enough information in the source material to assess the number of test cases, the full set of competing models, scoring variability or how criteria were weighted. The reported figures therefore describe results under the company’s evaluation, rather than a general measure of financial research performance.
It is also unclear how the results translate to live deals, where documents can be incomplete, inconsistent or updated over time. Hebbia did not quantify time saved, report how often human reviewers found errors, or provide client measurements of effects on decisions. Any expectation of less overnight work or faster review should be treated as a prospective benefit, not a reported deployment outcome.
The model was identified as a prerelease build. The material does not specify its broader availability, whether performance may change in a released version, or when the proposed monitoring agents might be developed. No independent evaluation is included.
Evidence Needed From Live Work
The next evidence that could clarify the practical value would include broader evaluations, fuller details about the test design and results from finance professionals using the system in actual research workflows. Reports on how reviewers verify generated answers and presentations, how often they identify errors, and whether the process changes research time would help distinguish benchmark performance from operational impact.
Hebbia has also described agents that could monitor information and flag developments for finance professionals. The source material gives no release timeline or performance data for that proposed role. Until further results or deployment details are available, the reported GPT-6 Astra figures remain company-run test results for selected financial research tasks.
Key Questions
What did Hebbia test GPT-6 Astra on?
Hebbia tested a prerelease build on source selection and presentation tasks in its Matrix platform, using financial research cases and company-defined grading criteria.
How did GPT-6 Astra perform in Hebbia’s tests?
Hebbia said the model scored 13.6 percentage points above its best tested alternative on required source types and met 97.9% of applicable presentation requirements, 14.4 percentage points above the best alternative on that measure. The company also reported that 99.5% of cited excerpts were relevant to the question.
Were these results independently verified?
The results described are from Hebbia’s own tests. The source material does not report an independent evaluation or provide enough test-design detail to judge how broadly the results apply.
Does this show GPT-6 Astra is ready for live deal decisions?
No. The tests do not establish performance in live deal work, and Hebbia did not report measured effects on decisions, error rates in review or actual time savings. Professionals would still need to check sources and conclusions.
When will Hebbia’s proposed research-monitoring agents be available?
The source material provides no release date or evaluation results for the proposed agents that would monitor information and surface signals to finance professionals.
Primary source: OpenAI · via ThorstenMeyerAI.com
