OpenAI launches ChatGPT for Financial Services with GPT-6 Astra, built-in market data and tools for producing research, financial models and client materials. The tailored ChatGPT Work product is initially aimed at investment banking and equity research teams at eligible financial institutions.

 

Morgan Stanley and Evercore helped shape the product as design partners. The September 10 launch combines included datasets, connections to firms’ existing subscriptions, reusable document templates and enterprise governance controls in one environment; OpenAI has not published standard pricing.

 

OpenAI Launches ChatGPT for Financial Services With Built-In Data

The central product decision is to include selected premium financial datasets rather than make every customer negotiate access and configure a connector first. OpenAI says data from Daloopa, PitchBook, LSEG News and Crunchbase is indexed and hosted on its infrastructure to improve retrieval, latency and source-level citations.

 

The launch groups financial information into three access paths:

  • Included premium datasets available inside the product
  • Entitlement links to a firm’s existing subscriptions
  • MCP connections to additional provider services

 

The included sources cover areas such as earnings transcripts, financial statements, company fundamentals and private-company data. OpenAI also names Quartr among the available providers in its launch materials, while Reuters independently reported the product’s release and its initial data roster.

 

For subscriptions firms already hold, OpenAI says it is working with S&P Capital IQ, LSEG, MSCI, Dow Jones Factiva and Moody’s on shared sign-in and entitlement integrations. The aim is for providers to recognize what a user is licensed to access through the person’s ChatGPT sign-in.

 

OpenAI has also optimized connections for services including S&P Global and FactSet. It says the wider connector ecosystem exceeds 50 integrations and includes Datasite, Box, Preqin and Intapp, although the exact data available will still depend on a firm’s contracts and administrator settings.

 

Morgan Stanley and Evercore Shape Banking Workflows

Morgan Stanley and Evercore served as design partners, focusing the first version on work performed by bankers and equity researchers. OpenAI says the collaboration identified two recurring constraints: reliable access to specialist data and the effort required to turn analysis into polished, firm-specific deliverables.

 

The official announcement highlights value analysis, leveraged-buyout modeling, buyer screening, earnings analysis and pitchbook preparation. Those examples place the product inside existing professional workflows rather than presenting it as a consumer investing assistant.

 

Administrators can publish approved Excel, Word and PowerPoint templates through a dedicated control page. Teams can then use the firm’s formats and style guides when producing valuation models, research notes and pitchbooks, reducing the gap between a generated analysis and material that can enter an internal review process.

 

Templates create consistency, but they do not establish that the numbers are correct. A valuation can be neatly formatted while relying on a stale period, an inappropriate peer set or a misunderstood accounting adjustment. Reviewers still need to trace inputs, inspect formulas and challenge the economic assumptions behind an output.

 

Related Coverage

 

GPT-6 Astra Connects Research to Financial Artifacts

GPT-6 Astra supplies the reasoning layer. OpenAI positions the model around three capabilities for finance: retrieving information from specialist sources, reasoning over financial material and generating artifacts such as documents, spreadsheets and presentations from the resulting analysis.

 

The company says the model can navigate figures, tables and supporting notes, compare values across periods and interpret annotations in public filings. Granular citations are intended to connect a number or claim to a specific table or passage, allowing a reviewer to inspect the evidence during the analysis rather than after a report is finished.

 

On OpenAI’s OfficeQA Pro benchmark, which uses U.S. Treasury Bulletins containing complex tables, charts and footnotes, GPT-6 Astra scored 69.9%, compared with 60.2% for GPT-5.6 Sol. That reported gain is relevant to document retrieval, but it is not a guarantee of accuracy on a bank’s private data or every modeling task.

 

Citations also have a defined boundary. They can show which source supplied a figure, but they cannot decide whether that source is current, whether an adjustment is appropriate or whether a conclusion satisfies a firm’s investment, legal or compliance standards. Those decisions remain accountable professional work.

 

Enterprise Controls Address Confidential Financial Work

ChatGPT for Financial Services builds on enterprise controls including SAML single sign-on, SCIM provisioning and role-based access. OpenAI says business data is not used to train its models by default, is encrypted at rest and in transit, and can be governed through configurable workspace-retention settings.

 

Compliance teams can export supported workspace logs through the OpenAI Compliance Platform for audits and investigations. Administrators can manage access to skills and apps by role, restrict supported read and write actions, and create multiple workspaces to maintain information barriers between teams.

 

Those controls are especially relevant when work involves material non-public information or client confidentiality. Institutions still need to map the product to their own data-classification, supervision, recordkeeping and model-risk policies. Availability of a control does not mean it has been configured for a particular regulatory obligation.

 

The service is available to eligible financial institutions through OpenAI’s sales process, with no public list price in the announcement. Expansion beyond investment banking and equity research is planned, while broader data coverage and future model updates are expected to enter the product over time.

 

The launch’s most consequential test will be whether institutions can reproduce cited analyses, audit generated models and measure time saved without weakening review. If those operational checks hold, specialized data access and firm templates may matter as much as the underlying model in determining adoption.