🔍 Read the full analysis: Could GPT-6 Astra Improve Financial Research? Hebbia Sees Progress on ThorstenMeyerAI.com
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
Hebbia says its August and September 2026 tests found a prerelease GPT-6 Astra performed better than the strongest alternative it tested on source coverage and following presentation requirements. The company’s results cover selected financial research cases, not live deals or broader model performance, and do not establish time savings or accuracy in deployment.
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 tasks in its financial research platform: finding relevant supporting sources and producing presentations that follow a user’s brief. The company said its August and September 2026 tests used financial research cases in Matrix, but the findings are company-run results and do not show how the model performs across other workflows or in live deals.
For source selection, Hebbia reported that GPT-6 Astra scored 13.6 percentage points above the best alternative it tested on citing required types of sources. The company also said 99.5% of cited excerpts were relevant to the question. For presentation tasks, Astra met 97.9% of applicable requirements, a 14.4-percentage-point lead over the best alternative on that measure.
Hebbia says it tested the models on the same financial research cases using the same grading criteria. Domain experts wrote or audited those criteria against the underlying documents, according to the company. The evaluation focused on selecting documents or Matrix cells that support an answer and generating native PowerPoint slides that match specified structures and guidelines.
Matrix brings together material such as financial statements, contracts and management presentations, and can link answers and slide footnotes to source documents. Hebbia’s example describes a banker reviewing roughly 2,000 files for a potential acquisition, examining matters such as customer concentration, contract terms and churn, then preparing a diligence summary and valuation presentation. The example is hypothetical, not a reported 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.
Why Source Links Matter in Deal Research
Financial research often requires professionals to connect evidence scattered across many documents, then present conclusions in a form colleagues or clients can review. Hebbia’s reported gains address two distinct parts of that process: locating relevant evidence and turning analysis into a presentation that follows a brief. Better performance on both could reduce some manual searching and editing, if the results hold in practical use.
The reported citation relevance may also help reviewers inspect where an answer came from. But a link does not establish that a conclusion is correct: a reviewer still needs to check whether a passage supports the claim and whether the figures use the right period or definition. The tests do not show that professionals can rely on generated analysis without review, or that it changes deal decisions.
The comparison is useful as a snapshot of Hebbia’s evaluation, not an independent benchmark. The figures concern the cases and criteria selected by the company. Broader testing would be needed to establish whether the results generalize to different firms, document sets, models or transaction workflows.
Products Worth Considering

Jackery HomePower 3000 Portable Power Station, 3072Wh

DJI Mini 4K Camera Drone Combo with 2 Batteries

Apple AirPods Pro (2nd Generation) Wireless Earbuds with USB-C MagSafe Charging Case
As an affiliate, we earn on qualifying purchases.
How Matrix Tests Financial Work
Hebbia’s Matrix platform is designed to organize research across source files and connect answers to supporting material. In the company’s acquisition scenario, a user could examine documents in a data room alongside outside information such as peer-company filings and industry news. Matrix could then help draft a diligence summary and slides using a firm’s template.
The reported evaluation focused on source selection and presentation compliance, rather than the full range of work involved in an acquisition. Hebbia says users can open citations to inspect the source document and highlighted passage, while slide footnotes can link to supporting material. That traceability is intended to help professionals verify the evidence before using an answer.
Hebbia also described a possible future role for agents that monitor information and surface signals for finance professionals. That proposal is separate from the reported model tests: the company gave no deployment timeline or performance results for such a feature.
“GPT-6 Astra scored 13.6 percentage points above the best alternative tested on citing required types of sources.”
— Hebbia
Limits of Hebbia’s Evaluation
Hebbia has not provided enough detail in the supplied account to assess the number of test cases, the full list of competing models, scoring variability or how criteria were weighted. The reported percentages therefore describe performance under the company’s evaluation, rather than a general measure of financial research accuracy.
It is also unclear how the results translate to live transactions, where files can be incomplete, inconsistent or updated during review. Hebbia did not report measured time saved, error rates found by human reviewers, or effects on client decisions. Any suggestion that the system reduces overnight work or speeds review remains an expectation, not a demonstrated deployment outcome.
The model is described as a prerelease build, and the source material does not specify its availability or release status. No independent evaluation is included. The proposed monitoring-agent use case has no stated launch date or reported test results.
Evidence Needed Beyond Test Cases
Hebbia says it intends to build on the results to help finance professionals assess deals using research that can be traced to its sources. Evidence that would clarify the practical value includes broader evaluations, more information about the test design, and results from use in actual financial workflows.
For any deployment, relevant follow-up measures would include how often reviewers correct generated answers, whether citations support the conclusions attached to them, and whether presentations require less editing. Hebbia has not reported those outcomes. Its idea of agents that monitor information and flag developments remains forward-looking, with no release timeline or performance evidence provided.
Key Questions
What did Hebbia say GPT-6 Astra did better?
Hebbia said the prerelease model outperformed the best alternative it tested on source selection and meeting requirements for generated presentations in its financial research cases.
What results did Hebbia report?
The company reported a 13.6-percentage-point lead on citing required types of sources, 99.5% relevance for cited excerpts, and a 97.9% rate for meeting applicable presentation requirements. These are results from Hebbia’s own tests.
Do the tests show that GPT-6 Astra is ready for live deals?
No. The supplied results do not establish performance in live transactions, and Hebbia did not report deployment outcomes, human-review error rates or effects on deal decisions.
Was the example involving 2,000 files a real acquisition?
No. Hebbia described it as a hypothetical example of how a banker might use Matrix to review documents and prepare diligence materials.
When will the reported monitoring agents be available?
Hebbia gave 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
