🔍 Read the full analysis: Exploring GPT-6 Astra’s Role In Hebbia’s Financial Research 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 of a prerelease GPT-6 Astra found higher scores than the best alternative it tested for identifying relevant sources and meeting presentation requirements. The results are company-run evaluations, not independent evidence of accuracy or time savings in live deals.
Hebbia says a prerelease version of OpenAI’s GPT-6 Astra outperformed the best alternative the company tested on two financial research tasks: selecting relevant sources and producing presentations that follow a user’s brief. The company reported the results from tests conducted in August and September 2026 using financial research cases in its Matrix platform. The findings are company-run and do not establish how the model performs across other tasks or in live transactions.
Hebbia evaluated models on source selection and slide creation, using what it described as the same financial research cases and grading criteria for each. 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 at citing required types of sources. Hebbia also reported that 99.5% of the excerpts cited by the model were relevant to the question.
For presentation tasks, Hebbia said GPT-6 Astra met 97.9% of applicable requirements, 14.4 percentage points above the best alternative on that measure. The tasks involved producing native PowerPoint slides that followed specified structures and guidelines. Matrix can link answers and slide footnotes to underlying materials, allowing users to inspect the cited document and passage, according to the company.
Hebbia illustrated the workflow with a hypothetical banker reviewing roughly 2,000 files for a potential acquisition. The platform could help examine customer concentration, contract terms and churn, incorporate peer filings and industry news, and draft a diligence summary and valuation presentation. The company did not describe this as a completed client deal or provide results from a live deployment.
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 reported scores address two connected demands in financial work: locating evidence across large document sets and communicating findings in a format that follows a brief. A higher rate of relevant source selection could make it easier for a reviewer to trace claims to documents. A draft that meets more presentation requirements could also reduce some editing, if the results carry over to actual work.
Those possible benefits depend on the analysis being sound, not just well-cited or neatly presented. Users still need to check whether a source supports the conclusion, whether figures use the right reporting period, and whether important evidence was missed. Hebbia did not report measured time savings, error rates found during review, or effects on client decisions. Its test results therefore describe performance under the company’s evaluation, not proof that the system can replace professional review.
The distinction between the two tasks matters in practice. A polished presentation can still be misleading if it omits relevant evidence; a well-supported answer may still require substantial work before it is ready for a client or decision-maker. Hebbia says GPT-6 Astra ranked first among the models it tested on both measures, but independent evaluations would be needed to assess whether that result generalizes.
Products Worth Considering

Amazon Kindle Scribe (16GB) with Premium Pen, Tungsten

Jackery Fridge Guard 1024Wh Backup Power Station with Mount

Bose QuietComfort Ultra Headphones (2nd Gen), Wireless Noise-Cancelling Over-Ear Headphones, Driftwood Sand
As an affiliate, we earn on qualifying purchases.
How Hebbia Tested Matrix
Matrix is Hebbia’s platform for bringing research together from documents and linking generated answers to their sources. The workflow described by the company can draw on materials such as financial statements, contracts and management presentations, alongside outside information including peer filings and industry news. Users can inspect citations and highlighted passages, and slide footnotes can link to source material.
Hebbia’s evaluation focused on whether a model selected documents or Matrix cells that supported an answer and whether it generated slides that followed specified requirements. The acquisition scenario involving an industrial-software company was presented as an illustrative example, not evidence of work on a named transaction. The reported findings concern a prerelease GPT-6 Astra build and the cases and criteria Hebbia selected for its tests.
“The company reported that GPT-6 Astra scored 13.6 percentage points above the best alternative tested on citing required types of sources.”
— Hebbia
Limits of the Company Tests
Hebbia’s account does not provide the number of test cases, full list of competing models, scoring variability or details on how criteria were weighted. It is also unclear how the scores would change with different financial research tasks, document collections or model versions. The reported figures should not be treated as a general accuracy measure for financial research.
The company did not quantify time saved, disclose error rates identified by human reviewers, or report measured effects on decisions in live deals. Real transaction files can be incomplete, inconsistent or updated over time, conditions not addressed in the reported results. Hebbia’s expectation that the tools could reduce overnight work or speed review is not a reported deployment outcome.
Hebbia also discussed a possible future role for agents that monitor information and surface signals to finance professionals. The company did not provide a release date, detailed capabilities or evaluation results for that idea. The source material identifies the tested model as prerelease; broader availability and performance outside Hebbia’s tests remain unclear.
Evidence Needed Beyond Tests
The next useful evidence would include broader evaluations with clear case counts, model comparisons and scoring methods, followed by results from use in real financial research workflows. Reporting on how professionals check generated answers and presentations—and what errors they find—would help readers judge whether the scores translate into dependable work or measurable time savings.
Hebbia has described developing tools to help finance professionals assess deals through research traceable to its sources. It has also raised the possibility of agents that monitor information and flag developments. No timeline or performance data for those agents was provided, so that remains a forward-looking proposal rather than a confirmed product milestone.
Key Questions
What did Hebbia report about GPT-6 Astra?
Hebbia said a prerelease GPT-6 Astra scored above the best alternative it tested on source selection and presentation requirements in financial research cases.
How did GPT-6 Astra score on the reported tasks?
Hebbia reported a 13.6-percentage-point lead on citing required types of sources, 99.5% relevance among cited excerpts, and a 97.9% rate for meeting applicable presentation requirements. These are results from the company’s own tests.
Were the results independently verified or measured in live deals?
The source material describes company-run tests and does not provide independent verification or results from completed live deals. Hebbia also did not report measured time savings or decision outcomes.
What does Matrix do in this workflow?
Matrix brings together information from research documents and can link answers and slide footnotes to source materials, which users can inspect. Hebbia describes it as supporting document review and presentation drafting.
When will Hebbia’s proposed monitoring agents be available?
Hebbia did not give a release date or report evaluation results for agents that monitor information and surface signals. Their timing and capabilities remain unknown.
Primary source: OpenAI · via ThorstenMeyerAI.com
