AI News
All stories
Reported

Hebbia reports better financial research results with prerelease GPT-6 Astra

Hebbia says its prerelease tests of GPT-6 Astra found gains in source coverage, citation relevance and presentation quality for financial research.

Products · · 2 min read

Hebbia reports that prerelease tests of OpenAI's GPT-6 Astra beat the best alternative it tried on two parts of financial research: finding the sources a question requires, and building a presentation that follows a banker's brief. OpenAI published the results as a customer story. The figures come from Hebbia's own tests.

The source-finding result is stated in two numbers. On citing the required types of sources, GPT-6 Astra scored 13.6 percentage points above the best alternative tested. Hebbia also reports that 99.5 percent of the excerpts it cited were relevant to the question. Those citations matter for Hebbia's product, Matrix, which links each answer to the documents behind it. A percentage-point lead is measured against the other model's score. The gap shows how far ahead the model was, but not either model's absolute score.

The presentation result is similar. Hebbia says GPT-6 Astra met 97.9 percent of the applicable presentation requirements, 14.4 percentage points ahead of the best alternative on that measure. The model is also described as producing native PowerPoint slides that follow the user's structure and guidelines. Kavita Anand, a member of Hebbia's technical staff on product, said that earlier models tended to excel at either craft or instruction following but not both. In her words, GPT-6 Astra is "number one on both of those fronts." Hebbia reports that it ranked first on both measures.

The method is described only in part. Hebbia tested a prerelease build in August and September 2026, using the same financial research cases and grading criteria for each model. Domain experts wrote or audited those criteria against source documents. The post does not say how many cases were run, which alternative models were tested, or how the presentation requirements were scored. Because Hebbia ran the tests and sells a product built on the model, readers have to rely on its description of the criteria.

The scenario Hebbia uses to explain the work is hypothetical. In it, a banker weighing an acquisition of an industrial-software company, with roughly 2,000 files to review, loads the data room into Matrix. The banker examines customer concentration, contract terms and churn in cited cells, then turns the research into a diligence summary and valuation deck in the firm's template. Anand said she does not want a deck that looks like AI made it. Hebbia wants bankers to spend less time reformatting decks and checking sources. She also described a future in which agents monitor information in the background and surface signals for finance professionals to assess. That is a forward-looking view from Hebbia, not a result from the tests.

What changed

Hebbia's own prerelease tests found GPT-6 Astra scored 13.6 percentage points above the best alternative tested on citing required source types and 14.4 points ahead on presentation requirements.

Type: New capability · Applies to: GPT-6 Astra

Why it matters

Firms that use AI for financial research may find citation accuracy and source coverage the most useful measures of a model. This account reports gains on both, though only from one customer's prerelease tests.

What you can do

Teams doing financial research could run their own source-coverage and citation checks on Astra before changing their workflows.

Original source

Where we saw it

1 source

OpenAI

HebbiaOpenAI

AI wrote this summary from the sources listed, and nobody reviewed it individually before publication. Check the original source before you rely on it. How it works