ChatGPT is becoming a workstation for finance professionals

OpenAI launches ChatGPT for Financial Services, a specialized version that combines GPT-6 Astra, licensed financial data, detailed citations, and internal document templates.

Finding a figure in an appendix, checking how adjusted EBITDA was calculated, comparing several companies, then turning the analysis into a presentation that follows the bank’s standards. OpenAI wants to bring these operations into a single environment with ChatGPT for Financial Services.

Introduced on September 10, 2026, the new service is a version of ChatGPT Work tailored to financial institutions. It combines GPT-6 Astra, integrated professional data, subscriptions already held by the organization, and tools for creating financial models, documents, and presentations.

The product initially targets investment banking and equity research. OpenAI highlights use cases including valuation analysis, LBO modeling, buyer screening, earnings analysis, and pitchbook preparation.

ChatGPT for Financial Services was not designed solely around scenarios imagined by OpenAI. Morgan Stanley and Evercore served as design partners, helping identify the most time-consuming tasks, the most frequently used sources, and the formats expected by financial teams.

These partnerships do not mean that either firm has transferred all its operations to the service. OpenAI says they helped shape the product and its initial use cases. The company has not disclosed how many employees are involved, how much time has been saved, or what share of their documents are produced with the platform.

The main difference from a general-purpose version of ChatGPT begins with the data. A model may understand the principles of valuation, but it still needs recent, detailed, and properly attributed figures to work on a real company.

OpenAI directly integrates information supplied by Daloopa, PitchBook, LSEG News, and Crunchbase. These datasets cover earnings releases, financial statements, company fundamentals, private companies, funding rounds, investors, acquisitions, and other areas.

Eligible institutions can use them without negotiating a separate contract with each provider or configuring an additional connection. The data is indexed and hosted on OpenAI’s infrastructure to improve retrieval, presentation, and its association with the model’s responses.

This setup distinguishes the product from a chatbot that merely searches the web. A financial figure may depend on a footnote, an accounting method, an adjustment, or a specific reporting period. A value on its own is not always enough to understand what it represents.

When normalizing a P&L statement, for example, ChatGPT can retrieve the reconciliation used to calculate adjusted EBITDA, display the related notes, and identify the expenses that were excluded. The user remains responsible for deciding whether those adjustments are relevant to the valuation.

Responses can include citations pointing to a specific table or passage. The interface highlights the supporting information so analysts can check the figure and its context without manually searching through the entire document.

Citations reduce the risk of receiving a number that cannot be traced, but they do not guarantee that the interpretation is correct. A genuine source may be associated with the wrong reporting period, another entity within the group, or a metric calculated under a different definition. Human review therefore remains necessary, particularly when the result will be shared with a client or used in a transaction.

The included datasets do not replace every database already purchased by financial institutions. Many banks hold contracts covering more specialized information, particular historical records, or usage rights negotiated for their teams.

OpenAI is therefore developing integrations with S&P Capital IQ, LSEG, MSCI, Dow Jones Factiva, and Moody’s. These providers will be able to recognize a user’s entitlements through their ChatGPT sign-in and grant access to content already covered by the institution’s subscriptions.

The service consequently separates two categories. Some data is provided directly with ChatGPT for Financial Services. Other information remains tied to the institution’s existing contracts and requires the provider to confirm the connected user’s access rights.

This distinction matters for both compliance and billing. The appearance of a provider’s name in the announcement does not necessarily mean that its entire catalog is included in the price of the service. Exact coverage will depend on the source, the rights held by the institution, and the integrations enabled by its administrator.

These agreements are supplemented by an ecosystem of more than 50 connections, including Datasite, Box, Preqin, FactSet, and Intapp. They make it possible to combine external information with the documents, data rooms, and tools already used by the institution.

OpenAI says it has optimized certain connections built on MCP, the protocol that allows a model to query an external service or database. The goal is to reduce configuration errors and failures when a request must travel across several systems.

That promise will need to be tested in real working environments. Permissions can vary between employees, providers may impose their own limits, and internal documents may use structures that are difficult to interpret. The announcement provides no connection failure rate, availability measurement, or comparison with a standard MCP setup.

Reasoning and response generation are handled in part by GPT-6 Astra. The model can navigate financial documents, interpret their tables and notes, conduct a multistep analysis, and then convert the result into a document, spreadsheet, or presentation.

ChatGPT for Financial Services is not permanently tied to this particular model. OpenAI plans to add future models as they are released, allowing institutions to gain new capabilities without rebuilding the entire product.

Continuous upgrades also introduce a challenge. A new model may change how responses are phrased, how sources are cited, or which assumptions are selected in a calculation. Regulated organizations will therefore need