Two finance SKUs shipped in four days. They do not buy the same job. OpenAI launched ChatGPT for Financial Services on 10 September 2026 as a tailored ChatGPT Work experience for eligible institutions, with design partners Morgan Stanley and Evercore. Anthropic launched Claude for Financial Advisors on 14 September 2026 (Reuters) as a Claude Cowork plugin for registered investment advisors.
Read them as a data-path choice. OpenAI indexes selected market data on its own infrastructure. Anthropic connects to the custodian, CRM, and portfolio stack the advisor already runs. A bank research desk and an RIA do not share a shortlist just because both headlines say “finance.”
Where the numbers live
- OpenAI: built-in datasets it says it indexes and hosts, including Daloopa, PitchBook, LSEG News, and Crunchbase. The launch page also quotes Fiscal.ai as a native data partner.
- OpenAI: working with S&P Capital IQ, LSEG, MSCI, Dow Jones Factiva, and Moody’s on shared sign-in and entitlement so existing seats can appear inside ChatGPT.
- Anthropic: connectors into Addepar, BlackRock Advisor Center, Schwab Advisor Services, Envestnet (Tamarac and MoneyGuide), iCapital, Orion (including Redtail), SS&C Black Diamond, Wealthbox, Wealth.com, Vanguard, and Zocks.
- Anthropic plugin setup bundle: BlackRock, Schwab, Addepar, Envestnet, iCapital, Orion, Wealthbox, Wealth.com, and Zocks. The advisor picks which to connect.
What OpenAI actually shipped
The product sits on ChatGPT Enterprise controls: SAML SSO, SCIM, role-based access. OpenAI says business data is not used for training by default, and that firms can create multiple workspaces to enforce information barriers. Admins can publish Excel, Word, and PowerPoint templates. Availability is for eligible financial institutions through sales, not a self-serve consumer plan. Reuters covered the launch. Terms live on OpenAI’s Financial Services Terms page.
The model is GPT-6 Astra. On the launch page OpenAI reports OfficeQA Pro scores of 69.9 percent for Astra and 60.2 percent for GPT-5.6 Sol on U.S. Treasury Bulletin tasks. Those are company-reported bench numbers. Keep them next to your own filings, not in place of them.
What Anthropic actually shipped
Claude for Financial Advisors is a plugin plus skills. Anthropic says the skills cover advisor onboarding, alternative-investment briefs, compliance and AI-policy checks, estate and tax briefs, portfolio rebalance review, post-meeting notes, pre-meeting prep, and prospect intake. Investment recommendations, client communications, and compliance determinations stay subject to human review and approval.
- Install path: Cowork plugin browser, then guided connector setup.
- Anthropic recommends Enterprise for RIAs because of audit logs.
- Firms that request a new license before the end of September 2026 receive a one-time usage credit.
- Some partners also ship their own plugins. The post names BlackRock Advisor Center, and existing S&P Global and LSEG plugins.
Anthropic cites Kitces for the claim that a typical practice spends only a sixth of its time in client meetings, and EBRI’s 2026 Retirement Confidence Survey for “over four in 10 American workers” who do not know who to go to for good financial or retirement planning advice. Those are Kitces and EBRI figures as quoted by Anthropic. They are not Anthropic measurements.
Approval and audit are not the same knob
OpenAI’s gate is eligibility, workspace barriers, and admin templates. Anthropic’s gate is human approval on regulated actions plus Enterprise logs. Pair either SKU with the gated cyber access map if the same firm also wants Daybreak, Mythos, or Fairwind. A wealth plugin is not a cyber seat. Fable 5.1 and Mythos 5.1 remain a separate access question.
Questions that make the two SKUs comparable
- Which client or research data is hosted by the vendor, and which only transits a connector?
- Which existing entitlements (S&P, LSEG, Factiva, Moody’s, Schwab, Addepar) actually work on day one, versus “working with”?
- Who can export a workspace or plugin transcript into the firm’s archive?
- What is blocked until a named human approves it?
- Is Enterprise mandatory for the audit story you will tell compliance?
- Does the contract allow the same SKU in a managed offering for your own clients?