OpenAI
As of August 2026. Vendor-published facts flagged; product facts on the quarterly re-verification list.
What it is
Model provider plus an agent build surface: the Agents SDK, reinforcement fine-tuning (GA May 2025) [vendor], and the enterprise agent platform positioning marketed as Frontier.
How it answers the bank
Evaluation and change (E). Reinforcement fine-tuning is the notable capability, and the research's verdict on it is a scoping one: it requires a programmable grader, which makes it the right tool for objectively gradable tasks and the wrong tool for style. That is a useful, honest boundary and it maps directly onto the guide's position that domain SMEs own the eval bar.
Interoperability (F). MCP support. Agents SDK is one of the frameworks in general use, in a market where only around 18% of instrumented organisations use any recognised framework at all.
Coverage
Core at R07 agent platform. Real at R06. Adjacent across the estate layers.
Watchouts
- This is the thinnest profile in the hub relative to the vendor's prominence, and that is a statement about available evidence rather than about the product. The enterprise platform positioning is not established in the research to the standard the question bank asks for: identity model, memory portability, metering transparency and governance surfaces are not documented in the sources this guide collected. Ask sections A, C and D directly and record the answers with dates.
- The two-estate problem applies in its sharpest form to any licensed enterprise platform: work running there does not flow through your gateway, and your telemetry and cost visibility depend on what the vendor exports.
Evidence status
The SDK and fine-tuning facts are vendor-published and documented. The enterprise platform's answers to the bank are not established here, and the guide declines to infer them from adjacent products.
Sources
research/R07, R06 vendors.md and sources.md.
Source: vendors/profiles/openai.md in the evidence repository behind this site.