ChatGPT Agent
by OpenAI
OpenAI's agentic mode in ChatGPT that browses, runs code, and acts
Last reviewed 2026-06-18
ChatGPT Agent is an agentic mode inside ChatGPT, launched July 2025, that lets ChatGPT take multi-step actions on its own virtual computer rather than only answering. It can navigate websites, filter results, prompt the user to log in securely when needed, run code in a terminal, conduct analysis, and produce editable deliverables like slide decks and spreadsheets. It connects to tools such as Gmail and GitHub and can use APIs to interact with other apps, fluidly switching between reasoning and action to complete a task end to end under the user's oversight. It unified OpenAI's earlier Operator (browsing) and Deep Research (analysis) capabilities into one agent, available to ChatGPT Pro, Plus, Team, Enterprise, and Edu users via agent mode. OpenAI reported a 41.6% pass@1 score on Humanity's Last Exam for the agent model; treat such figures as vendor-reported benchmarks.
What it can do
Navigate websites and act on a virtual computer
SupervisedUses its own virtual computer to browse and interact with sites, filtering results and prompting the user to log in securely when needed.
sourceRun code and conduct analysis
SupervisedExecutes code in a terminal tool and performs data analysis as part of completing a task.
sourceProduce editable deliverables
SupervisedGenerates editable slideshows and spreadsheets summarizing its findings.
sourceConnect to tools and apps
SupervisedConnects to tools like Gmail and GitHub and uses APIs to interact with other applications, with the user overseeing actions.
source
Strengths
- +Unifies browsing (Operator) and analysis (Deep Research) into one agent
- +Acts on a virtual computer: browses, runs code, and builds deliverables
- +Available across ChatGPT paid plans with no separate product to adopt
Limitations
- −Agentic browsing can be slow and error-prone on complex sites and login/CAPTCHA flows
- −Requires user oversight; not a hands-off autonomous worker
- −Benchmark figures are vendor-reported
Overview
ChatGPT Agent is an agentic mode inside ChatGPT, launched in July 2025, that lets ChatGPT take multi-step actions on its own virtual computer rather than only answering. It unified OpenAI's earlier Operator (browsing) and Deep Research (analysis) into a single agent.
What it does
It navigates websites, filters results, prompts for secure logins when needed, runs code in a terminal, conducts analysis, and produces editable slide decks and spreadsheets. It connects to tools like Gmail and GitHub and uses APIs to interact with other apps, switching between reasoning and action to complete tasks under the user's oversight.
Autonomy note
It is a supervised agent: it performs multi-step work but pauses for secure logins and confirmations, and OpenAI frames the user as overseeing its actions. Agentic browsing can be slow and unreliable on complex sites.
Integrations & setup
Available inside ChatGPT via agent mode for Plus, Pro, Team, Enterprise, and Edu users. Connects to Gmail, GitHub, and other apps via connectors.
Pricing
Included in paid ChatGPT plans rather than sold separately.
Best for / not for
Best for ChatGPT users who want the assistant to actually carry out research, analysis, and document creation. Less suited to high-stakes unattended automation where reliability and oversight matter.
Alternatives
OpenAI's Operator was its browsing predecessor (now folded in); Manus and Genspark are general agentic assistants; Gemini and Perplexity offer agentic/research modes.
What people are saying
We aggregate real LinkedIn discussion into sentiment for the agents people search most. ChatGPT Agent isn't tracked yet, want it added? Request tracking.
FAQ
How is ChatGPT Agent different from regular ChatGPT?+
ChatGPT Agent is an agent mode that takes multi-step actions on its own virtual computer (browsing, running code, building files) rather than only generating text, completing tasks end to end under the user's oversight.
Is ChatGPT Agent fully autonomous?+
No. It performs multi-step tasks but operates under user oversight, pausing for secure logins and confirmation on sensitive actions, so it is a supervised agent in practice.
Sources
- Introducing ChatGPT agent (OpenAI) · accessed 2026-06-18
- OpenAI launches a general purpose agent in ChatGPT (TechCrunch) · accessed 2026-06-18
- ChatGPT agent release notes (OpenAI Help Center) · accessed 2026-06-18
Last reviewed 2026-06-18