Editorial independence: AI Agent Square is not paid by the vendors we review. We currently earn no commissions from links on this site, and no vendor can pay to influence scores, rankings, or review content. How reviews are made is on our methodology page.
OpenAI's first web-browsing agent, launched as a research preview in January 2025. On OpenAI's own benchmarks it handled simple web tasks and failed most full desktop tasks.
| Plan / Tier | Cost | Availability |
|---|---|---|
| ChatGPT Pro | $200/month | Operator research preview, US only, from January 2025 |
| ChatGPT Pro, Plus, Team | Plan price | ChatGPT agent (Operator's successor) from 17 July 2025; Enterprise and Edu from August 2025; since withdrawn |
| computer-use-preview API (Developers) | $3 per 1M input tokens, $12 per 1M output tokens | OpenAI's computer-use model for developers, labelled a preview |
Operator worked through its own browser by typing, clicking and scrolling, so it did not need API cooperation from the target website.
Operator returned control to the user when a site needed a password or payment details, and paused for confirmation before completing major actions.
OpenAI reported a 38.1% success rate on OSWorld, a benchmark of full computer tasks. Most of those tasks failed.
Operator lasted about six months before it was folded into ChatGPT agent, and ChatGPT agent was itself withdrawn in 2026. Buyers who built processes on either had to move them.
Operator worked inside a web browser. Internal systems without a web interface were out of reach.
Operator ran on OpenAI's Computer-Using Agent (CUA) model, which combined GPT-4o's vision capabilities with reinforcement learning. It read screenshots of the page and acted with mouse and keyboard input rather than through site APIs.
OpenAI published these results for CUA at launch: 38.1% on OSWorld (full computer use), 58.1% on WebArena and 87% on WebVoyager (both browser tasks). OpenAI noted that most WebVoyager tasks are relatively simple.
Developers can use OpenAI's computer-use model as computer-use-preview at $3 per 1M input tokens and $12 per 1M output tokens, plus a per-call tool fee listed on OpenAI's pricing page.
OpenAI's current agent inside ChatGPT, launched in July 2026, which OpenAI points former ChatGPT agent users to for longer, multi-step tasks.
Anthropic's computer-use capability for Claude, offered through its API.
View Claude ReviewMicrosoft's agent builder, a fit for organizations already on Microsoft 365 and Power Platform.
View Copilot Studio ReviewOperator can no longer be bought. On OpenAI's own figures it handled simple web tasks and failed most full desktop tasks. Teams that want this capability now should evaluate ChatGPT Work or build on the computer-use API, and keep a human approving anything that spends money or submits data.
No. OpenAI integrated Operator into ChatGPT agent on 17 July 2025 and retired the standalone site. ChatGPT agent has since been withdrawn too, and OpenAI points users to ChatGPT Work.
OpenAI reported 38.1% on OSWorld, 58.1% on WebArena and 87% on WebVoyager for the CUA model behind Operator.
No. It returned control to the user for passwords and payment details and paused for confirmation before major actions.
OpenAI lists computer-use-preview at $3 per 1M input tokens and $12 per 1M output tokens, plus a per-call tool fee.