OpenAI's financial services push this month was less about a single product announcement and more about a strategy becoming visible. GPT-5.4, the model released alongside a new suite of finance-specific tools, has scored 87.3% on OpenAI's internal investment banking benchmark — up from 43.7% with its predecessor GPT-5 on tasks like building financial models with proper formatting, citations, and scenario analysis. That's not an incremental improvement; it's the difference between a capable assistant and one that can meaningfully compress the hours analysts spend on grunt work. Bloomberg's framing — "rivaling Anthropic" — signals that what had been a two-horse race for financial services AI is now clearly competitive.
The most practically significant move is ChatGPT for Excel, currently in beta. Embedding the model directly into the spreadsheet rather than requiring a separate interface changes the workflow in a subtle but important way: analysts don't have to context-switch out of the tool they're already using. The system can build, analyse, and update financial models using the existing formulas and structures a team already relies on, rather than generating outputs that then need to be formatted and transferred. This is the kind of integration friction reduction that typically drives actual adoption, as opposed to demos.
The financial data integrations launched alongside GPT-5.4 are equally telling. Moody's, Dow Jones Factiva, MSCI, Third Bridge, and MT Newswires are now accessible within ChatGPT, with FactSet coming soon. These are the data sources that investment bankers, credit analysts, and equity researchers actually use — not generic internet search. The pitch is that an analyst can pull market data, company filings, news context, and third-party research into a single workflow without switching between platforms or copying and pasting between data sources. If it works reliably in practice, it compresses a significant portion of the research preparation process.
The competitive context matters. Anthropic had established itself as the preferred AI partner for financial institutions that care about safety and reliability — Goldman Sachs, Deutsche Bank, and others had built integrations around Claude. OpenAI is now moving directly onto that turf with purpose-built financial tools and benchmark numbers designed to make the performance comparison explicit. Morgan Stanley, which has been among the more prominent OpenAI financial services partners, serves as a case study for how the tools are being evaluated against real workflows rather than synthetic tests.
BBVA, the Spanish banking giant, expanded its OpenAI collaboration this month, announcing it would make ChatGPT available to all 74,000 of its global employees and deploy it across customer-facing workflows. BBVA is a useful data point because it is large enough that deployment friction is real — rolling out AI tools to 74,000 people in multiple countries with different regulatory requirements is not a trivial exercise. If the rollout proceeds smoothly, it will be cited as evidence that enterprise-scale financial AI is past the pilot phase.
The competitive dynamic in financial AI is now settled enough to have clear dimensions. OpenAI brings raw capability and the ubiquity of the ChatGPT brand. Anthropic brings a reputation for reliability, detailed audit trails, and the willingness to accept contractual responsibility for model outputs that regulated institutions require. Neither has a decisive advantage across the full spectrum of financial use cases. What's becoming clear is that financial institutions are no longer choosing whether to adopt AI — they're choosing which AI, at what tier, for which workflows. That's a market structure OpenAI's GPT-5.4 push was explicitly designed to shape.