AIThis post was created with the assistance of artificial intelligence (AI).

🔍 Read the full analysis: Hebbia Sees Improvements In Financial Research With GPT-6 Astra on ThorstenMeyerAI.com

Buying for a business?Offer from Amazon

Get business pricing on tech for your team

  • Business-only prices and quantity discounts
  • Tax-exempt purchasing
  • Multiple users, one account, clear invoices
As an affiliate, we earn on qualifying purchases.

TL;DR

Hebbia says a prerelease build of OpenAI’s GPT-6 Astra scored above the best alternative it tested on source selection and presentation requirements in financial research cases. The results come from company-run tests 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 supporting sources and producing presentations that follow a user’s brief, as detailed in the original analysis. The company said its tests, conducted in August and September 2026 in its Matrix platform, used financial research cases and criteria reviewed by domain experts; the results are company-reported and do not show how the model performs across live deals or other workflows.

On source selection, Hebbia reported that GPT-6 Astra scored 13.6 percentage points above the best alternative it tested for citing required types of sources. The company also said 99.5% of cited excerpts were relevant to the question. On presentation tasks, the model met 97.9% of applicable requirements, a 14.4-percentage-point lead over the best alternative on that measure. Those figures describe results in Hebbia’s evaluation, not independently verified performance in general financial research.

Hebbia ran the tests in Matrix, its platform for bringing together research from documents and linking answers to their sources. The company said it used the same financial research cases and grading criteria for each model, and that domain experts wrote or audited the criteria against the underlying documents. The reported tasks covered choosing documents or Matrix cells to support answers and creating native PowerPoint slides that followed specified structures and guidelines.

To illustrate the workflow, Hebbia described a hypothetical banker reviewing about 2,000 files for a potential acquisition. Matrix could help examine customer concentration, contract terms and churn, draw on peer filings and industry news, and prepare a diligence summary and valuation presentation. The scenario is an example from the company, not a reported client deal. Hebbia says answers and slide footnotes can link back to documents so a reviewer can inspect the supporting passage or figure.

At a glance
reportWhen: Tests conducted in August and September…
The developmentHebbia reported that a prerelease GPT-6 Astra performed better than the other models it tested on two financial research tasks in Matrix.
Financial research · Briefing

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.

97.9%
Applicable presentation requirements met
13.6 points
Above the best alternative on source coverage
14.4 points
Above the best alternative on presentation requirements
99.5%
Cited excerpts relevant to the question

What Hebbia tested

Tests in August and September 2026 used financial research cases and the same grading criteria for each model.

Source selection

Find supporting evidence

GPT-6 Astra scored higher on citing required types of sources. Matrix links answers to documents for review.

Presentations

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.

1
Review deal documents

Examine roughly 2,000 files for details such as customer concentration, contract terms and churn.

2
Connect outside research

Bring in peer filings and industry news alongside the data room documents.

3
Draft and check

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.

Reported
  • Domain experts wrote or audited the criteria.
  • GPT-6 Astra ranked first among tested models on both measures.
Still unclear
  • How results transfer to messy, changing live deal documents.
  • Actual time saved, review error rates and independent replication.
Review neededA relevant excerpt can still be used incorrectly. Check the reporting period and whether the cited passage supports the conclusion.
The takeaway
Hebbia’s results are a narrow signal that GPT-6 Astra may help with evidence gathering and presentation drafting; broader evaluations and real-world review data are needed to judge its value in deal work.

How Source Links Could Affect Deal Review

The results address two distinct steps in document-heavy financial work: finding evidence across a large collection of files and presenting analysis in a requested format. If a model can help with both, it could reduce some of the work involved in gathering and organizing information for diligence. But Hebbia did not report measured time savings, review costs or effects on deal decisions, so the practical impact remains unquantified.

Traceable citations may make generated research easier to check, but a citation alone does not establish that a conclusion is correct. Reviewers still need to confirm that the passage supports the claim, that figures use the right reporting period and that documents are current. A polished presentation can also omit important evidence, while a well-supported answer may still need substantial editing before it is ready for a client or decision-maker.

For financial professionals considering AI tools, the reported comparison offers a narrow signal about how one platform evaluated a prerelease model on selected tasks. It is not evidence that users can delegate analysis or rely on generated material without professional review. Independent evaluations and results from actual deployments would help establish whether the gains extend beyond Hebbia’s test cases.

Products Worth Considering

Apple MacBook Air 15-inch Laptop with M5 Chip, 16GB Memory, 512GB SSD, Midnight

Apple MacBook Air 15-inch Laptop with M5 Chip, 16GB Memory, 512GB SSD, Midnight

As an affiliate, we earn on qualifying purchases.

Matrix Tests and the Deal Example

Matrix is designed to collect information from documents such as financial statements, contracts and management presentations, then connect answers to their supporting material. In Hebbia’s acquisition example, a user could combine files from a data room with outside sources such as peer-company filings and industry news. The platform could then help produce a diligence summary and slides based on a firm’s template.

The evaluation focused on source selection and presentation compliance, rather than establishing overall model accuracy across finance. Hebbia said the criteria were tied to the source documents, giving evaluators a basis to judge whether required evidence appeared and whether slides followed instructions. The company’s account also describes citations as part of the workflow: users can open links to inspect documents and highlighted passages.

The acquisition scenario involving an industrial-software company is hypothetical. Hebbia presented it to explain how the tools might be used, not as evidence that GPT-6 Astra completed a live transaction or produced a verified client outcome.

“99.5% of cited excerpts were relevant to the question.”

— Hebbia

Limits of Hebbia’s Evaluation

The information provided does not specify the number of test cases, the full list of competing models, variation in scores or how individual criteria were weighted. Without those details, readers cannot determine how robust the reported differences are or how well the cases represent financial research more broadly. The results also concern a prerelease model build, and the account does not establish whether the same performance would hold in a later release.

Hebbia did not report a measured rate of errors caught during review, time saved in use, or effects on client decisions. It is also unclear how the system would handle live deal materials that are incomplete, inconsistent or updated over time. The company’s statements about reducing overnight work or speeding review are expectations, not reported deployment outcomes. No independent evaluation or client results were included in the supplied account.

Hebbia has also described a possible future role for agents that monitor information and flag developments for finance professionals. The company has not provided a release date, details of the proposed capabilities or evaluation results for that idea.

Evidence Needed From Broader Use

The next evidence that could clarify the reported gains would include fuller details about Hebbia’s test design, evaluations on a broader range of financial cases and comparisons that can be independently assessed. Results from actual use could also show whether the model reduces review time, how often its output needs correction and whether citations help users identify unsupported conclusions.

For now, the reported findings are limited to Hebbia’s tests of a prerelease GPT-6 Astra in Matrix. The company has outlined a possible future direction in which agents monitor information and surface signals for finance professionals, but it has not provided a timeline or performance data for that proposed capability. The effect on live financial work remains to be established.

Key Questions

What did Hebbia report about GPT-6 Astra?

Hebbia said the prerelease model scored above the best alternative it tested on source selection and presentation requirements in financial research cases run in Matrix.

What were the reported test results?

Hebbia reported a 13.6-percentage-point lead on citing required types of sources and said 99.5% of cited excerpts were relevant. For presentation tasks, it reported that the model met 97.9% of applicable requirements, 14.4 percentage points above the best alternative on that measure.

Were the tests independent?

No independent evaluation was described. Hebbia conducted the tests and reported the results; it said domain experts wrote or audited its criteria against the source documents.

Does the report show that GPT-6 Astra saves time on live deals?

No. Hebbia did not report measured time savings, error rates in deployment or effects on deal decisions. The acquisition workflow in its account is hypothetical.

What remains unknown about the evaluation?

The supplied account does not give the number of cases, the full set of competing models, score variability or the weighting of criteria. It also does not establish how results translate to live work or to a later model release.

Primary source: OpenAI · via ThorstenMeyerAI.com

You May Also Like

Fed’s Waller On A Half-Point Hike: Possible Effects On Jobs And AI

Fed Governor Christopher Waller says a half-point December hike may be needed unless the Fed signals a January move. The decision is not settled.

Forezai · TradingAgents: A Trading Firm Made of Agents

Forezai introduces TradingAgents, a framework of specialized AI agents mimicking a trading desk, emphasizing structured disagreement and oversight.

The memory chip crunch is paying off for this U.S. company

Micron’s Q3 earnings quadrupled as the U.S. memory chip maker benefits from the global shortage driven by AI demand, with strong future forecasts.

Cloud’s Hidden Memory Bill

Memory shortages are causing hidden cost increases in cloud services, impacting prices for high-memory instances and prompting reconsideration of cloud strategies.