
Hebbia reports GPT-6 Astra gains in financial research

Hebbia, which builds Matrix for financial research, tested OpenAI’s GPT-6 Astra on two parts of a banker’s workflow: finding the sources a question requires and building a presentation that follows the brief. Before a banker can recommend an acquisition, they need to make sense of financial statements, customer contracts, and management presentations, often across thousands of files. In Hebbia’s example, a banker evaluating the acquisition of an industrial-software company with roughly 2,000 files could load the data room into Matrix and ask about revenue durability, peer comparisons, customer concentration, contract terms, and churn, with answers linked to cited cells. Matrix could bring in peer filings and industry news to help develop the deal thesis, then turn the research into a diligence summary and valuation deck in the firm’s template. In that workflow, GPT-6 Astra selects documents or Matrix cells to support an answer and converts the analysis into native PowerPoint slides that follow the user’s structure and guidelines. Kavita Anand, Member of Technical Staff, Product, at Hebbia, said the leaps GPT-6 Astra has made in output quality and accuracy have been surprising, and that it shines compared with any other model the company has tested on these two fronts.
Hebbia tested a prerelease build in August–September 2026, using the same financial research cases and grading criteria for each model; domain experts wrote or audited the criteria against source documents. GPT-6 Astra scored 13.6 percentage points above the best alternative tested on citing the required types of sources, and 99.5% of the excerpts it cited were relevant to the question. On presentation generation, it met 97.9% of the applicable requirements, 14.4 percentage points ahead of the best alternative. Anand said other models tended to trade off craft and instruction following, while GPT-6 Astra was the first model she had seen that ranked first on both fronts.
Traceability is central because users base multi-million-dollar decisions on what the AI returns. Citations let users open the underlying document and highlighted passage, and slide footnotes link back to source material; a reviewer could verify a customer-concentration figure against the source table and correct the reporting period before the deck goes to the investment committee. Anand said even a single percentage-point gain can yield substantial marginal benefit in a data room with thousands of documents, and that getting a deck closer to the final output can reduce overnight work to a few minutes of reviewing GPT-6 Astra’s output. Hebbia wants bankers to spend less time reformatting decks and checking sources, and more time assessing a deal.
Anand also sees a future role for agents that monitor information in the background and surface signals for finance professionals to assess, potentially letting users start thinking about opportunities before competitors. Hebbia wants to build on these results so finance professionals can assess deals with research traceable back to its sources.

