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

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

Before you orderOffer from Amazon

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
Start your free Prime trial Free trial for eligible customers · Cancel anytime
As an affiliate, we earn on qualifying purchases.

TL;DR

Hebbia says a prerelease build of OpenAI’s GPT-6 Astra outperformed the best alternative it tested on source coverage and presentation requirements in financial research cases. The company-run tests do not establish performance in live deals or other workflows, and Hebbia has not reported measured time savings or independent validation.

Hebbia says a prerelease version of OpenAI’s GPT-6 Astra outperformed the best alternative it tested on two financial research tasks: finding relevant sources and producing presentations that follow a banker’s instructions. The company reported the results from tests run in August and September 2026 in Matrix, its document research platform; the findings are company-reported and do not establish how the model performs in live deals or other workflows.

On source selection, GPT-6 Astra scored 13.6 percentage points above the best alternative Hebbia tested on citing required types of sources. Hebbia also said 99.5% of cited excerpts were relevant to the question. On presentation tasks, the model met 97.9% of applicable requirements, a result the company said was 14.4 percentage points higher than the best alternative on that measure.

Hebbia said it ran the models on the same financial research cases and applied the same grading criteria. Domain experts wrote or audited those criteria against the underlying documents, according to the company. The two tasks measured whether a model selected documents or Matrix cells that supported an answer, and whether it created native PowerPoint slides that followed a specified structure and guidelines.

The company’s example involves a banker reviewing roughly 2,000 files for a potential acquisition. Matrix could help examine customer concentration, contract terms and churn, bring in peer-company filings and industry news, and produce a diligence summary and valuation presentation. Hebbia says citations in answers and slide footnotes link back to source documents. The acquisition scenario is hypothetical, not a completed client deal.

At a glance
reportWhen: Tests conducted in August and September…
The developmentHebbia reported results from August and September 2026 tests in which prerelease GPT-6 Astra scored higher than the best alternative it tested on two financial research tasks.
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.

Source Checks and Deal Presentations

The results focus on two linked tasks in financial work: assembling evidence across large collections of documents and turning research into a presentation that meets a brief. In deal reviews, source coverage can affect whether an analysis accounts for relevant contracts, filings or customer data. A presentation that follows the requested structure may also need less editing before a human reviewer can use it.

But strong performance on these measures is not the same as verified financial accuracy or a reliable deal recommendation. A citation must support the claim attached to it, and figures still need to be checked for details such as the reporting period. Human review remains necessary, especially when documents conflict, omit information or have been updated.

Hebbia’s numbers offer a view of how one platform is evaluating AI for document-heavy finance work. They do not show that the model saves time in practice, reduces errors during live transactions or improves investment decisions. The reported gains are best read as results on selected cases and criteria, rather than proof of broad operational impact.

Products Worth Considering

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

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

As an affiliate, we earn on qualifying purchases.

How Matrix Handles Financial Research

Hebbia’s Matrix platform brings together information from files such as financial statements, contracts and management presentations. It can tie answers to supporting documents so users can inspect the cited material. In the company’s described workflow, a banker could combine data-room documents with outside sources such as peer filings and industry news, then use the research to prepare a diligence summary and slides.

The reported evaluation separated source selection from presentation compliance. That distinction matters: a polished deck may omit important evidence, while a well-supported answer may still require substantial work to become a client-ready presentation. Hebbia says GPT-6 Astra ranked ahead of the other models it tested on both measures, but the source material does not provide independent results to show whether that balance holds across other tests.

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

— Hebbia

Limits of Hebbia’s Test Results

Hebbia has not provided enough detail in the source material to assess the number of test cases, full list of competing models, scoring variability or weighting of criteria. The reported figures reflect the company’s evaluation, and there is no independent comparison establishing how the results generalize to other financial research tasks.

It is also unclear how the results translate to live deal work, where records can be incomplete, inconsistent or changing. Hebbia did not report measured time saved, the rate of errors caught in review or client data on effects on decisions. Statements about reducing overnight work or speeding review are expectations, not reported deployment outcomes.

The company has also described a possible future role for agents that monitor information and surface signals to finance professionals. No release date, deployment details or evaluation results for that proposed capability were provided.

Evidence Needed From Live Work

The next evidence that could clarify the practical value would include broader evaluations with more detail on the cases, comparison models and scoring methods, alongside results from use in actual financial workflows. Reporting on how professionals verify generated answers and presentations, and whether deployments change review time or error rates, would help distinguish test performance from operational impact.

Hebbia has said it wants to build on these results to help finance professionals assess deals using research traceable to its sources. Its proposed monitoring agents remain a forward-looking idea rather than a feature with a stated release date or reported performance. For now, the confirmed development is the company’s test report; broader adoption and real-world outcomes remain unestablished.

Key Questions

What did Hebbia test GPT-6 Astra on?

Hebbia tested whether the model could select sources relevant to financial research questions and create PowerPoint presentations that met specified structures and guidelines, using cases in its Matrix platform.

What results did Hebbia report?

The company said GPT-6 Astra scored 13.6 percentage points above the best alternative it tested on source coverage and met 97.9% of applicable presentation requirements, 14.4 percentage points above that alternative. Hebbia also reported that 99.5% of cited excerpts were relevant to the question.

Were the results independently verified?

The source material describes company-run tests and does not provide independent validation. Hebbia says domain experts wrote or audited the grading criteria against the underlying documents.

Does the report show that GPT-6 Astra saves bankers time?

No measured time savings or error rates were reported. Claims that the workflow could reduce review work are expectations, and the results do not establish effects in live deal work.

What is still unknown about the evaluation?

The source material does not state the number of cases, full competitor list, scoring variability or criteria weighting. It also does not show how the results generalize to other workflows or how the model performs when documents are incomplete or change over time.

Primary source: OpenAI · via ThorstenMeyerAI.com

You May Also Like

Unlock The Power Of AI In Wireless Earbuds: 2026’S Best Models

Explore the top wireless earbuds of 2026 with advanced AI features, balancing sound, battery, comfort, and noise cancellation for every need.

Exploring Astra’s AI Path: Critical Skills And Safety Protocols

OpenAI has published a page titled ‘Path to Astra,’ signaling efforts to develop advanced AI capabilities alongside safety measures, though details remain unclear.

What Stanford Professors’ Books Reveal About AI Training Practices

A headline attributed to The Stanford Daily alleges Anthropic used Stanford professors’ books to train Claude, but key evidence and legal details are unavailable.

Asana’s Browser Testing: 76X Lower Model Costs With GPT-6.1 Sol

A headline says Asana cut model costs 76x in browser tests using GPT-6.1 Sol, but provides no baseline, methodology or performance results.