ChatGPT enters investment banks with Astra model, integrating market data and valuation calculations in a closed institutional environment

OpenAI has launched a financial-sector version called ChatGPT for Financial Services, powered by the inference model GPT-6 Astra, in a move that aims to contain complex financial transactions within an institutional work environment that meets banking audit standards. Access to the service is currently limited to qualified financial institutions, with the first rollout focusing on the workflows of investment banks and equity research, after the product was developed with input and feedback from major financial institutions including Morgan Stanley and Evercore.

The new platform relies on a direct integration of financial data packages hosted by OpenAI and indexed on its own infrastructure, eliminating the need for analytical teams to negotiate separate contracts or build independent technical connections for each provider. The integrated data includes sources such as Dalupa, PitchBook, London Stock Exchange Group news and CrunchBase, covering earnings release texts, financial statements and data on public and private companies. The company is also working alongside S&P Capital IQ, MSCI, Dow Jones Factiva and Moody’s to integrate single sign-on and permission management mechanisms, and the system is compatible with more than 50 connectors, including Dataset, Box, Brekin and Entap.

On the technical side, the model was designed.GPT-6 Astrato read numbers, tables and explanatory footnotes attached to financial documents accurately, achieving a score of 69.9 % on the standard Office QA Pro competency test, surpassing its predecessor GPT-5.6 Sol, which scored 60.2 %. This capability allows bankers to break down earnings items before interest, taxes, depreciation and amortization adjustments and to examine excluded costs during calculation to assess their direct impact on valuation models. To support traditional work environments, the platform lets supervisors upload internally approved Excel, Word and PowerPoint templates to create research memos and investment marketing presentations in line with each institution’s strict formatting standards.

The detailed citation feature occupies a central position in the system, allowing every number and claim to be traced to its original source for verification during analysis. In terms of security and governance, the platform inherits the controls of ChatGPT for Enterprise, including single sign-on via the SAML protocol, account management through SCIM, role-based permission settings, and the ability to isolate multiple workspaces to maintain strict information barriers within the bank. Data are encrypted in transit and at rest, with a publicly stated commitment not to use institutional data to train models virtually and with supervisors granted the ability to set retention policies, while the company has not disclosed service pricing or precise qualification criteria.

What changes in practice in the regional work environmentThis shift places investment-bank teams and private-equity departments in the Gulf and Egypt in front of a new operational reality. Instead of junior analysts spending hours extracting earnings calls and reconciling valuation tables across disparate data platforms, the required skill shifts to quickly auditing the model’s assumptions and reviewing the adjusted earnings items it proposes. In turn, risk and compliance committees at regional banks and investment funds will face a careful trade-off between the efficiency of centrally accessing global market data through a unified cloud infrastructure and the stringent regulatory requirements to keep transaction books and proprietary pricing models confidential within institutional boundaries.