🔍 Read the full analysis: 3 Ways GPT-6 Astra Is Helping Hebbia’s 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 prerelease GPT-6 Astra outperformed the best alternative it tested on source coverage and presentation requirements in financial research tasks. The August and September 2026 tests were company-run and do not establish performance in live deals or other workflows.
Hebbia says a prerelease build of OpenAI’s GPT-6 Astra outperformed the best alternative it tested on two financial research tasks: identifying relevant source material and creating presentations that follow a brief, according to the original analysis. The company’s tests, conducted in August and September 2026 in its Matrix platform, measured performance on selected cases and do not establish how the model performs across live deals or other workflows.
On source selection, GPT-6 Astra scored 13.6 percentage points above the best alternative tested at citing required types of sources, Hebbia reported. The company also said 99.5% of cited excerpts were relevant to the question, a focus shared by other GPT-6 Astra evaluation work. These are separate measures: the first concerns coverage of required source types, while the second reflects the relevance of excerpts that were cited.
For presentation tasks, Hebbia said GPT-6 Astra met 97.9% of applicable requirements, a result 14.4 percentage points above the best alternative on that measure. The tests assessed whether the model could produce native PowerPoint slides that followed specified structures and guidelines. Hebbia said it used the same financial research cases and grading criteria for each model, with domain experts writing or auditing the criteria against the underlying documents.
The evaluation took place in Matrix, Hebbia’s platform for gathering research from documents and linking answers to supporting material, an approach also relevant to research on GPT-6 Astra. In a company-described example, a banker reviewing roughly 2,000 files for a potential acquisition could examine customer concentration, contract terms and churn, then draw on peer filings and industry news to prepare a diligence summary and valuation presentation. Hebbia described that scenario as an example, not 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.
Evidence and Slides in Deal Research
The results address two connected demands in financial research: finding evidence across large document sets and turning analysis into a presentation that meets a specific brief. In deal work, source coverage and presentation quality can affect how much additional checking and editing a first draft requires. A polished deck can still be inadequate if it omits required evidence; a well-supported answer may still need substantial work before it is ready to present.
Hebbia says Matrix citations can link answers and slide footnotes back to documents, allowing users to inspect the source passages and figures. That traceability may help reviewers check how a conclusion was produced. It does not, by itself, show that the conclusion is accurate or that the cited evidence supports the interpretation. Professionals would still need to verify details such as the reporting period behind a customer-concentration figure and whether a passage justifies the claim made.
The figures are relevant as an account of how one platform vendor is testing AI for document-heavy finance work. They are not evidence that users can rely on generated analysis without review, or that the model will reduce time or errors in a live transaction. Hebbia did not report measured time savings, review error rates or effects on deal decisions.
Products Worth Considering

Jackery HomePower 3000 Portable Power Station, 3072Wh

Bose QuietComfort True Wireless Earbuds with Active Noise Cancellation, Bluetooth 5.3, USB-C Charging, IPX4, Black

LG UltraGear 27-inch QHD OLED Gaming Monitor, 240Hz, 0.03ms, G-SYNC Compatible, FreeSync Premium Pro, HDR True Black 400
As an affiliate, we earn on qualifying purchases.
How Matrix Handles Deal Documents
Matrix is designed to bring together materials such as financial statements, contracts and management presentations, with answers linked to source documents. In Hebbia’s acquisition scenario, a user could work across files in a data room and add external information such as peer-company filings and industry news. The platform could then assist with a diligence summary and slides using a firm’s template.
The evaluation focused on two capabilities: selecting documents or Matrix cells to support an answer, and producing slides that follow instructions. Hebbia says users can open citations to inspect the document and highlighted passage, and that slide footnotes can link back to source material. The tests therefore assessed both evidence selection and presentation compliance, rather than establishing broad financial accuracy across all tasks.
“99.5% of cited excerpts were relevant to the question.”
— Hebbia
Limits of the Company Tests
The results come from Hebbia’s own evaluation of a prerelease model build. The material provided does not specify the number of test cases, the complete set of competing models, scoring variability or how individual criteria were weighted. Without those details or independent evaluation, it is difficult to judge how broadly the reported margins apply.
It is also unclear how the results would translate to live deal work, where documents can be incomplete, inconsistent or updated over time. Hebbia did not report measured time savings, error rates found during human review, or client outcomes. Its suggestions that the workflow could reduce overnight work or speed review are expectations, not reported deployment findings.
Hebbia has also discussed the possibility of agents monitoring information and flagging signals for finance professionals. The company has not provided a deployment timeline, described the proposed capabilities in detail or reported evaluation results for that future role.
Evidence Needed From Live Work
The next useful evidence would include broader evaluations with more detail about test cases, scoring and competing models, alongside results from real financial research workflows. Reporting how professionals check generated answers and presentations—and how often they find unsupported or incorrect material—would help clarify the practical value beyond benchmark-style scores.
For now, Hebbia’s reported results describe performance under its own test conditions, not a demonstrated effect on deal timelines or decisions. The company has outlined background monitoring by agents as a possible future direction, but has not given a release date or performance data for that proposal.
Key Questions
What did Hebbia test GPT-6 Astra on?
Hebbia tested a prerelease GPT-6 Astra build on source selection and presentation tasks in its Matrix platform, using financial research cases.
What results did Hebbia report?
The company said GPT-6 Astra scored 13.6 percentage points above the best alternative on citing required types of sources, with 99.5% of cited excerpts relevant. It also said the model met 97.9% of applicable presentation requirements, 14.4 points above the best alternative on that measure.
Do the results show GPT-6 Astra works in live deals?
No. The reported tests were company-run evaluations on selected cases. Hebbia did not report results from live deals, client outcomes or measured effects on financial decisions.
Can users verify the model’s answers?
Hebbia says Matrix links answers and slide footnotes to source documents, allowing users to inspect supporting passages. Users would still need to check whether the source is relevant and supports the conclusion.
When will finance agents that monitor information be available?
Hebbia has described monitoring agents as a possible future role, but has not provided a release timeline or reported evaluation results for that proposed capability.
Primary source: OpenAI · via ThorstenMeyerAI.com
