Hebbia tested OpenAI's GPT-6 Astra on two core financial-research tasks-finding and citing relevant sources, then building presentation decks-and posted double-digit gains over the best alternative models. The results point to AI that can cut hours of overnight document review down to minutes of verification, a shift with direct consequences for dealmakers who base multimillion-dollar decisions on what the AI returns.
What Hebbia tested
Hebbia's Matrix platform pulls together the documents a banker needs to assess a deal: financial statements, customer contracts, management presentations, peer filings and industry news. The company ran GPT-6 Astra through the same financial-research cases and grading criteria it uses for every model it evaluates. Domain experts wrote or audited the criteria against the source documents. The tests used a prerelease build between August and September 2026.
Two jobs were measured. First, selecting the right documents or data cells to answer a question. Second, turning that analysis into native PowerPoint slides that follow the user's structure, branding and formatting rules.
Citations and source coverage
GPT-6 Astra scored 13.6 percentage points above the best alternative tested on citing the required types of sources. When Hebbia checked the excerpts the model cited, 99.5% were relevant to the question asked.
"Traceability is really important to our users because they're basing millions of dollars of decisions on what the AI actually comes back with," said Kavita Anand, Member of Technical Staff, Product at Hebbia. "Citations help with traceability as well as giving users confidence that the AI found the right answer."
In a typical workflow, a banker might load roughly 2,000 files from a data room into Matrix to examine customer concentration, contract terms and churn. Each answer links back to a specific cell or document passage. A reviewer can open the citation, see the underlying source and highlighted excerpt, and verify a figure before the deck reaches the investment committee.
Presentation quality and following the brief
GPT-6 Astra met 97.9% of the applicable presentation requirements, putting it 14.4 percentage points ahead of the best alternative on that measure. Hebbia also ranked it first on both polish and adherence to the brief-a combination the company had not seen before.
"In the past, we've seen models that really shine on one or the other of these dimensions where they're really good at craft, but the instruction following and adhering to guidelines is not strong, or vice versa," Anand said. "GPT-6 Astra is the first model we've seen where it's number one on both of those fronts."
For bankers, that means a first draft that looks like work they would send to a client, in their firm's branding, with slide footnotes that trace back to source material. Anand described the practical impact: "The overnighters that you have to pull get reduced to a few minutes of reviewing GPT-6 Astra's output."
A future role for agents
Anand also pointed to a future where agents monitor information in the background and surface signals for finance professionals. "Imagine getting an important signal a few hours before your competitors," she said. "You can get started thinking about it and see and process more opportunities that a human is not manually hunting."
Hebbia plans to build on these test results to help users assess deals with research they can trace back to its sources. The company has been a partner with OpenAI across its lifetime, and Anand described the leap in output quality and accuracy as "truly surprising."
Why this matters for finance and sales professionals
When a model can cite the right sources 13.6 points more often and hit 97.9% of presentation requirements, the time saved is not theoretical-it flows straight to the bottom of a deal team's week. For sales and finance professionals, that means less manual reformatting, faster verification of key figures, and more capacity to assess the deal itself rather than the deck. The marginal gain Anand described matters most in data rooms with thousands of documents, where even a single point of accuracy can change a recommendation. Professionals who want to build the skills to evaluate and deploy these tools can explore AI for Finance Courses or Generative AI Courses that cover the models and workflows reshaping the industry.
