OpenAI launches ChatGPT for Financial Services with GPT-6 Astra

For the past two years, junior analysts on Wall Street have been caught in an awkward dance: using ChatGPT to summarize notes while constantly worrying about hallucinated metrics and compliance red flags.

OpenAI took that problem head-on by launching ChatGPT for Financial Services, a specialized workspace purpose-built for investment banking and equity research.

Developed alongside design partners Morgan Stanley and Evercore, the setup runs on the new GPT-6 Astra model and directly tackles the two biggest headaches in corporate finance: accessing verified financial datasets and generating presentation-ready deliverables.

Native financial feeds with verifiable citations

Instead of making corporate IT teams struggle with external MCP connectors or messy API keys, OpenAI simply licensed and hosted data natively from providers like Daloopa, PitchBook, LSEG News, and Crunchbase.

Analysts can jump straight into public filings, earnings transcripts, and private company records without separate contracts. Because OpenAI indexes this data internally, retrieval latency drops significantly.

More importantly, the system supports granular citations. When an analyst runs a P&L normalization analysis and inspects adjusted EBITDA, they can click every single metric to trace it back to the exact footnoted table in the source document.

For firms that already hold enterprise subscriptions to S&P Capital IQ, LSEG, MSCI, Moody's, or Dow Jones Factiva, shared sign-in allows immediate entitlement mapping. There are also over 50 optimized connectors for platforms like Datasite, Preqin, Box, and FactSet.

Custom pitchbooks and audit-grade controls

Financial work lives and dies in Excel and PowerPoint. GPT-6 Astra was tailored across three core pillars: retrieval, financial reasoning, and artifact synthesis.

Admins can upload proprietary corporate templates through an admin portal. Rather than spitting out generic tables, the model turns complex LBO modeling or earnings research directly into spreadsheets and pitchbooks that match the bank's exact typography and layout guidelines.

On security, customer data is never used to train OpenAI models by default. The platform includes SAML SSO, role-based controls, information barriers to prevent internal leaks, and full compliance log exports.

Access is currently restricted to eligible financial institutions, requiring teams to contact OpenAI sales directly.

Why it matters

This feels like OpenAI's most aggressive push yet to capture Wall Street's lucrative software budgets.

For investment analysts, this could eliminate the mind-numbing junior grind. Sifting through hundreds of footnote lines, reconciling line items, and pasting clean numbers into pitchbooks takes up half an analyst's week. Compressing that into minutes fundamentally reshapes deal prep, though human sign-off will remain non-negotiable given the legal stakes of financial filings.

For incumbent financial terminal providers, the warning lights should be flashing. Financial data used to be locked behind proprietary terminal interfaces that cost tens of thousands per seat. By indexing raw data directly and layering frontier reasoning on top, OpenAI turns static numbers into actionable models on demand.

OpenAI played this smartly by partnering with Morgan Stanley and Evercore instead of building in a vacuum. The real hurdle now is institutional trust: getting risk-averse compliance committees comfortable with allowing frontier models inside live M&A deal rooms.