On September 10, OpenAI released ChatGPT for Financial Services, a version of its enterprise product aimed squarely at investment banking and equity research. Built on GPT-6 Astra, it ships with premium data already inside: earnings call transcripts, financial statements, and private market information. The goal is to keep research, financial modeling, and client-ready deliverables inside a single workspace. Morgan Stanley and Evercore took part in the design.
Removing the work of bringing data in
The first thing that stands out is not the model but the roster of data sources included by default. Daloopa, PitchBook, LSEG News, Crunchbase, and Fiscal.ai are built in, covering earnings transcripts, financial statements, company fundamentals, and private market funding activity without a separate contract.
The more interesting part is how existing subscriptions are handled. For S&P Capital IQ, LSEG, MSCI, Dow Jones Factiva, and Moody's, the product recognizes what the institution already licenses and opens only the entitled material. At firms where database contracts differ from desk to desk, not having to police that boundary by hand matters more than it sounds.
Internal systems connect through MCP connectors, with more than 50 available. Datasite for deal management, Box for file storage, Preqin for alternative investments, FactSet for market data, and Intapp for business operations have optimized implementations.
What GPT-6 Astra is expected to handle
The underlying model is positioned as strong at three things: pulling the right numbers out of large document sets, reasoning through financial logic, and producing finished artifacts. Those artifacts are not limited to prose; spreadsheets and presentations count as well.
On OfficeQA Pro, a benchmark that measures work with complex financial documents, GPT-6 Astra scored 69.9 percent against 60.2 percent for the previous GPT-5.6 Sol. The gap is roughly 10 points on paper, but in documents with tangled cross-references, such as earnings materials and prospectuses, that margin often decides whether output is usable as is or has to be rebuilt.
There is also machinery for keeping output consistent. Excel, Word, and PowerPoint templates can be distributed and managed across the firm, so generated material does not drift with individual taste. Detailed citations show where each figure came from. In finance, being able to explain why a number is what it is often matters more than the number itself, which makes this a practical requirement rather than a nicety.
Whether it survives an audit decides adoption
When financial institutions evaluate tools like this, governance is what settles the question. ChatGPT for Financial Services supports SAML SSO and SCIM provisioning, role-based access control, and does not train on business data by default. Encryption at rest and in transit, along with configurable workspace retention, are also included.
Audit logs are handled through the compliance platform. Firms can additionally stand up multiple workspaces to create information barriers between departments. The regulatory requirement that investment banking and research must not share information is expressed directly in the product structure.
Aimed at the work junior staff have carried
Coverage of the launch has framed the product as taking on tasks traditionally handled by junior bankers: company research, financial analysis, and pitchbook preparation, the work that has long defined the hours in that job.
That said, what has been disclosed so far stops at features and benchmarks. There is no operational data showing how the product performed on live mandates. Pricing is undisclosed, availability is limited to eligible financial institutions, and access runs through OpenAI sales teams. Whether the same results appear outside Morgan Stanley and Evercore, the two design partners, remains to be seen.
The contest moves to bundled data
OpenAI is not alone in targeting finance. Anthropic offers Claude for Financial Services with agents built around specific tasks such as document preparation, audit readiness, and anti-money-laundering investigations. Google Cloud began offering Gemini Enterprise for Financial Services on August 25, with more than 50 skills aimed at automating research work.
Lined up side by side, the competitive axis is shifting away from raw model performance toward how much of the path to financial data, and the permissions around it, a vendor can absorb into the product. From the buyer's side, the choice narrows to whichever option demands less effort to reconnect licensed databases and less effort to convince the control functions. OpenAI putting automatic entitlement recognition front and center is an attempt to eliminate that friction before it is raised.
Summary
On September 10, OpenAI released ChatGPT for Financial Services, built on GPT-6 Astra. It embeds data from Daloopa, PitchBook, LSEG News and others, automatically recognizes existing S&P Capital IQ and Moody's subscriptions, and connects to internal systems through more than 50 MCP connectors. It scored 69.9 percent on OfficeQA Pro and supports SSO, role-based access control, and information barriers through separated workspaces. Pricing is undisclosed and availability is limited to eligible financial institutions. Morgan Stanley and Evercore participated in the design.
