OpenAI's finance tier of ChatGPT claims built-in financial data without saying where it comes from

The phrase doing the most work is "built-in financial data." Before a figure from it lands in a model that lands in a client deck, an analyst has to know which vendor supplied the series and as of when. On dev.to's reading of the announcement, OpenAI does not describe how the financial data is sourced, updated or presented inside the product ^[13].

Treat the phrase the way you would treat a benchmark table. It is a claim about someone else's inputs. For it to carry into a comp set or a covenant test, the series behind it has to be the one your team already reconciles against, and no series is named ^[13].

The dev.to write-up does the counting. It pulls three clear claims out of the announcement: the experience is tailored for financial services teams, it includes built-in financial data, and it uses GPT-6 Astra's reasoning alongside that data ^[3]. It sets those against six functions the announcement does not confirm ^[12] and five categories of buying question it leaves blank ^[11]. Counting both lists, that is eleven enumerated unknowns set against the three capabilities the announcement asserts ^[14].

What an announcement leaves out may still be in the product. dev.to describes what it had as a high-level product overview, not a full product specification ^[10]. So the read stays narrow: a team evaluating this now has no published answer to the questions its own approval workflow asks first.

The three stated uses sit in sequence. Research informs a model, and model outputs shape a client presentation or other customized material ^[6]. Putting all three in one environment takes out the handoffs, and it also takes out the places where somebody used to re-key a number and notice it was wrong. The announcement sets out no review process, so each organization has to define one itself ^[7].

Reasoning quality is not the constraint that decides this in a regulated finance function. A reviewer who cannot date a figure sends the deliverable back whatever produced it, and the three claims OpenAI makes here are all claims about capability ^[3].