ChatGPT agent
ChatGPT agent merges Operator and deep research into one system on its own virtual computer; its first model rated High biological risk.
One agentic model shifts between a visual browser, a text browser, a terminal and API connectors to finish multi-step tasks, producing slides and spreadsheets. Operator was folded in and its standalone site retired. OpenAI activated its High-capability biological safeguards precautionarily, without definitive evidence of novice uplift.
- Date
- Thursday, 17 July 2025
- Lab
- OpenAI
- Kind
- product
- Access
- app only
- Price
- Pro, Plus and Team at launch
Figures
| Measure | Value | Measured by |
|---|---|---|
| Humanity's Last Exam (pass@1) | 41.6% company-reported SOTA at launch | company |
| FrontierMath (with tool use) | 27.4% OpenAI-elicited, graded by Epoch AI; OpenAI has exclusive access to most private Tier 1-3 questions | mixed |
| BrowseComp | 68.9% +17.4 points over deep research | company |
| SpreadsheetBench (direct spreadsheet editing) | 45.5% | company |
The system card (2025-07-17) treated agent as High capability in Biological and Chemical under the Preparedness Framework as a precaution, which became the default posture for later frontier models.
Sources
This record was checked against its sources on 6 October 2026. How we check