Overview
- ChatGPT Agent merges the capabilities of Operator and Deep Research into a single AI that uses its own virtual computer to complete complex, multi-step tasks like web browsing, code execution and calendar management.
- The feature is available immediately to US Pro, Plus and Team subscribers, with rollout to Enterprise and Education tiers scheduled later this summer and EEA and Swiss access pending.
- OpenAI’s new reinforcement-trained model behind the agent posts state-of-the-art benchmark scores but may take 15–30 minutes to finish more challenging operations.
- The agent runs in a sandbox that asks for user approval before irreversible actions, employs Watch Mode on sensitive tasks and restricts financial and high-risk operations.
- Separately, OpenAI is piloting an integrated in-chat payment and checkout system with partners such as Shopify to earn commissions on product recommendations.