Where OpenAI Operator Stands in May 2026
OpenAI Operator was the company's first dedicated computer-using agent product, launched as a research preview in January 2025. By August 31, 2025, the standalone Operator surface was shut down and its capabilities were absorbed into ChatGPT Agent, the rebranded successor that runs inside the main ChatGPT product surface. As of May 2026, "Operator" as a product no longer exists; the underlying Computer-Using Agent (CUA) model and its capabilities live on inside ChatGPT Agent and are exposed to developers through the OpenAI Agents SDK. This page tracks the evolution timeline, current state, and competitive landscape.
Operator Timeline (Jan 2025 to May 2026)
| Date | Milestone |
|---|---|
| 2025-01-23 | Operator launched as research preview, ChatGPT Pro only, powered by the new CUA model (GPT-4o vision + RL on GUI interaction) |
| 2025-03 | Expansion to Plus and Team tier users in selected regions |
| 2025-07-17 | ChatGPT Agent announced, combining Operator + Deep Research into a single surface inside ChatGPT |
| 2025-08-31 | Standalone Operator product shut down; capabilities folded into ChatGPT Agent |
| 2025-Q4 | CUA exposed in API as "computer-use" tool for the Agents SDK |
| 2026-04-15 | OpenAI Agents SDK update with enterprise safety controls (TechCrunch report) |
| 2026-05 (current) | ChatGPT Agent active on Plus, Pro, Business, and Enterprise; CUA available via API to developers |
Computer-Using Agent (CUA) Benchmark Scores
| Benchmark | Score | Notes |
|---|---|---|
| OSWorld (full computer use) | 38.1% | State-of-the-art at launch; still leads at May 2026 |
| WebArena (web tasks) | 58.1% | Mid-tier reliability on complex multi-step flows |
| WebVoyager (visual web) | 87.0% | Strong on visually navigable sites; weak on CAPTCHA / JS-heavy |
The 38.1 percent OSWorld score on full computer use sets the ceiling for current general-purpose agentic browsing. The shutdown of standalone Operator was partially driven by reliability gaps on complex flows (purchases, CAPTCHA, session management) that the OSWorld and WebArena numbers reflect.
Why Operator Was Shut Down (and Folded Into ChatGPT Agent)
OpenAI publicly framed the August 31, 2025 shutdown as a consolidation, but the practical reasons were threefold:
- Reliability gaps on commerce flows. Operator failed to reliably complete purchases on websites with complex JavaScript checkout flows, CAPTCHAs, and session-management quirks. The product's headline use case (book a flight, order a product) had a sub-50 percent success rate on real-world sites.
- Better integration value inside ChatGPT. ChatGPT Agent could share context, memory, and personalisation with the rest of the ChatGPT product, removing the "copy-paste from chat to Operator" friction.
- Anthropic Computer Use competitive pressure. Anthropic shipped Computer Use in October 2024, with stronger integration into the developer-facing API (CUA was initially product-only). OpenAI's product folded; the API approach won.
Current OpenAI Agentic Product Surface (May 2026)
| Product / Surface | What It Does | Access |
|---|---|---|
| ChatGPT Agent | Combined web/computer agent for end users | Plus, Pro, Business, Enterprise inside ChatGPT |
| OpenAI Agents SDK | Multi-agent framework for developers | Open-source, PyPI / npm |
| computer-use tool (API) | CUA model exposed as a tool for custom agents | OpenAI API |
| Custom GPTs | Lightweight no-code agents | ChatGPT Plus+ |
Competitive Landscape
| Competitor | Vendor | Notable Capability |
|---|---|---|
| Anthropic Computer Use | Anthropic | API-first, strongest developer adoption |
| Perplexity Comet | Perplexity | Browser-native agentic experience |
| ChatGPT Atlas | OpenAI (separate from Operator) | Dedicated agentic browser, paired with ChatGPT Agent |
| Browser Use (OSS) | Open source | 94K GitHub stars; most-deployed open agentic-browser framework |
| Apple Intelligence Actions | Apple | iOS / macOS app automation |
What This Means for AI Visibility
Agentic browsing is no longer a niche capability; it is the substrate of an emerging class of buyer interactions where the human delegates the entire purchase or research flow to an AI agent. Brands now need to be discoverable, accurately representable, and compatible with structured-checkout flows inside computer-using agent stacks. ChatGPT Agent (the Operator successor) is one of three primary surfaces along with Anthropic Computer Use and Perplexity Comet, with Apple Intelligence Actions becoming meaningful through 2026. For brand-visibility programmes, Operator-era tracking should now be redirected to ChatGPT Agent + the underlying CUA model in API-based agent stacks.
Methodology
Timeline and capability data aggregated May 14, 2026 from OpenAI's official blog (Operator announcement, CUA technical post, ChatGPT Agent announcement), TechCrunch April 2026 Agents SDK coverage, and CUA benchmark scores from OpenAI's published evaluation. Operator's shutdown date and rationale from OpenAI's deprecation notice and third-party coverage. Refreshed monthly while the agentic surface is evolving rapidly.
How Presenc AI Helps
Presenc AI tracks brand-mention rates inside ChatGPT Agent flows alongside the rest of the ChatGPT surface, with specific instrumentation for agentic-purchase scenarios that traditional chat-only tracking misses. As the Operator-to-Agent consolidation completes and CUA capabilities spread into the developer-API ecosystem, brand-visibility programmes need to extend beyond chat-prompt tracking to include agent-tool-call tracking, which is where commercial decisions are increasingly made.