OpenAI launches a ChatGPT built only for financial firms

Who helped build it says a great deal about who it is aimed at. OpenAI said it refined the product through a design partnership with Morgan Stanley and Evercore, starting with investment banking and equity research. The company's account is that the sorest points for those teams were reliable access to data and the quality of the artifacts they produced.

The biggest change is in how the data is attached. Paid data from providers such as Daloopa, PitchBook and LSEG News ships with the product, so there are no separate contracts to negotiate and no connectors to stand up. OpenAI said it indexes and hosts this data on its own infrastructure, in order to improve retrieval accuracy and latency and to support a feature that traces a figure back to the table it came from.

The companies supplying the data have signed on to that arrangement. Thomas Li, chief executive of Daloopa, said, "We are excited to be selected as one of the native data partners for OpenAI for Financial Services," adding that it "continues our strategy of being the data infrastructure for AI and agentic workflows in financial services." Tom Van Buskirk, executive vice president of technology and engineering at PitchBook, explained that "the private markets have always been difficult to see into, and getting a clear picture means piecing together data scattered across the market."

There is a separate path for firms that already hold data subscriptions. OpenAI said it is working with S&P Capital IQ, LSEG, MSCI, Dow Jones Factiva and Moody's on shared sign-in and entitlement integrations. When a user signs in with a ChatGPT account, the provider recognizes them and automatically opens the data they are already entitled to.

The firms plugging in have put their names to it as well. Sally Moore, chief client officer and co-head of market intelligence for Kensho Data & Platforms at S&P Global, said, "Trusted data is the foundation every AI workflow is built on." Emily Prince, group head of enterprise AI at LSEG, said the focus is on making the company's licensed content usable through MCP connectivity and a curated news selection. There are more than 50 connectors, and the company wrote that heavily used MCPs such as S&P Global and FactSet have been tuned for immediate use.

Putting data in and reading it out are two different jobs. OpenAI offered the example of an analyst normalizing a profit-and-loss statement, opening the reconciliation and notes behind an adjusted EBITDA figure to see which costs were excluded and then judging whether to use that number in a valuation. The company said its work with partners feeds into post-training, product improvements and expansion into other financial categories, and wrote that it will teach its models to handle this data the way the best analysts do.

The performance claims come with numbers. OfficeQA Pro, which OpenAI cited, tests whether an agent can move across the complex tables, charts and footnotes in U.S. Treasury Bulletins and find what it needs; GPT-6 Astra scored 69.9% against 60.2% for GPT-5.6 Sol. The company argued the model is state of the art on three fronts: information retrieval, financial reasoning and artifact generation.

Turning analysis into actual work product is built into the tool as well. Administrators can publish Excel, Word and PowerPoint templates from a dedicated page, after which teams can produce valuation models, research notes and pitchbooks in the firm's own format and style guide. The workflows the company listed include value analysis, LBO modelling, buyer screening, earnings analysis and pitchbook preparation.

Security is the first thing a financial firm asks about. The product sits on ChatGPT Enterprise's SAML single sign-on, SCIM provisioning and role-based access controls, and firms can split into multiple workspaces to enforce information barriers. OpenAI said business data is not used to train its models by default, that administrators set retention, and that compliance teams can export workspace logs through its compliance platform into existing audit workflows.

What this announcement changes is not a list of tools but the order in which a banker opens the day. Until now an analyst opened a data terminal, a document viewer and a spreadsheet separately and moved between them by hand. Once paid data sits inside the chat window, that order inverts and a single window becomes the entrance to every other tool. For the companies selling data, the place where they sell moves from their own terminal screen into the inside of someone else's product.

This is a fight over position rather than price. OpenAI itself wrote that this one product cannot meet the finance industry's needs and that a range of solutions will be required, and it limited availability to eligible financial institutions. METAL reported on OpenAI's ChatGPT agent that searches a company's own material, and this financial-services product lays paid data and industry templates on top of that structure.

METAL read the announcement in full, and not one price appears in it. What does appear, sentence after sentence, is who indexes the data and who holds the sign-in. The next thing a financial firm has to negotiate is not a model's score but which infrastructure its own data sits on.