OpenAI Operator Review: What Happened to It (and What Replaced It)
Quick Answer: OpenAI Operator is no longer available. It launched in January 2025 as a research-preview browser agent, was folded into “ChatGPT agent” in July 2025, and was fully shut down on August 31, 2025. ChatGPT agent has since been retired too as of this writing, OpenAI directs users to ChatGPT Work, a broader agent built on GPT-5.6 that launched July 9, 2026. If you’re searching for “OpenAI Operator” today, ChatGPT Work is the product you actually want.
Operator (Jan-Aug 2025): OpenAI first standalone browser-use agent. Pro-only, U.S.-only, research preview.
ChatGPT agent (Jul 2025-2026): Merged Operator’s browser skills with Deep Research’s analysis skills into one agent mode inside ChatGPT.
ChatGPT Work (Jul 2026-present): The current product. Built on GPT-5.6, folds in Codex, adds file/doc creation, Scheduled Tasks, connectors to 1,400+ apps, and a built in desktop browser.
Pricing today: Free ($0), Go ($8/month), Plus ($20/month), Pro (from $100/month), plus Business/Enterprise/Edu tiers.
No official benchmark score (OSWorld/WebArena style) has been published for ChatGPT Work yet we’re flagging that gap rather than guessing.
Key Takeaways
1. Operator was a genuine product, not a rumour it existed, worked, and was OpenAI first real “browser agent.”
2. It didn’t fail so much as get absorbed twice. Operator → ChatGPT agent → ChatGPT Work is a lineage, not a dead end.
3. The current product does more than Operator ever did: it creates finished documents, connects to your actual work tools, and can run unattended for hours.
4. If your goal was “an AI that can use a browser for me,” that capability is alive and well it just doesn’t carry the Operator name anymore.
Entity Definitions
Operator OpenAI original browser use AI agent, powered by a model called Computer-Using Agent (CUA), which combined GPT-4o’s vision with reinforcement-learning trained reasoning to click, type, and scroll on real webpages.
ChatGPT agent The successor that combined Operator’s web interaction with OpenAI Deep Research tool, launched as an “agent mode” inside ChatGPT itself rather than a separate site.
ChatGPT Work The current agent, built on GPT-5.6, that goes further: it can produce finished spreadsheets, slides, and documents, work across connected business tools, and keep projects running in the background for hours.
The Full Timeline: Operator’s Story
January 23, 2025 Operator launches
OpenAI introduced Operator as a research preview: an agent with its own browser that could look at a webpage and interact with it by typing, clicking, and scrolling no custom API integration required. It launched exclusively for ChatGPT Pro subscribers in the United States, with plans to expand to Plus, Team, and Enterprise later.
Under the hood, Operator ran on Computer Using Agent (CUA), a model combining GPT-4o’s vision capabilities with reasoning trained via reinforcement learning. In OpenAI own published benchmarks at launch, it scored 38.1% on OSWorld (general OS level tasks) and 58.1% on WebArena (web-specific tasks) solid for a first-generation agent, but well short of human-level accuracy, and OpenAI was upfront that it struggled with complex interfaces and long workflows.
July 17, 2025 Operator becomes “ChatGPT agent”
Six months in, OpenAI merged Operator with Deep Research, its information-synthesis tool, into a single “ChatGPT agent” mode accessible directly inside ChatGPT composer for Pro, Plus, and Team users. OpenAI own explanation: many tasks people tried with Operator were actually better suited to Deep Research, so they combined both into one system rather than maintaining two separate tools.
This is the point where the standalone Operator site (operator.chatgpt.com) began winding down OpenAI original announcement page still carries the update notice today.
August 31, 2025 Operator is shut down
The standalone Operator product was fully discontinued. From this date, “agent mode” inside ChatGPT was the only way to get browser-agent functionality from OpenAI.
2026 ChatGPT agent is also retired
By the time of this review, OpenAI own Help Centre states plainly: “ChatGPT agent is no longer available.” It’s a short-lived successor OpenAI moved fast again, this time replacing agent mode with a considerably more capable product.
July 9, 2026 ChatGPT Work launches
This is where things land today. ChatGPT Work is described by OpenAI as an agent that can “take action across your apps and files, stay with a project for hours if needed, and turn a goal into finished work.” It’s powered by GPT-5.6 (in Sol, Terra, and Luna variants) and absorbs Codex OpenAI coding agent, which by this point had more than 5 million weekly users, over 1 million of them already using it for non-coding work.
Feature Breakdown: What ChatGPT Work Actually Does
| Capability | Detail |
|---|---|
| Document creation | Produces polished slides, sheets, docs, and web apps (“Sites,” in public beta) from gathered context |
| App connectors | 1,400+ plugins Slack, Microsoft Teams, Google Drive, SharePoint, Salesforce, Gmail, and more |
| Scheduled Tasks | One-time or recurring automated runs e.g., reviewing new Slack messages weekly and updating a doc |
| Built-in browser (desktop) | Lets ChatGPT gather information, use web tools, and refine work without leaving the app |
| Computer Use (desktop) | Executes tasks across your apps and files in the background clicking, typing, moving files |
| Plan mode | ChatGPT proposes a step-by-step plan before starting work, which you can adjust or approve |
| Cross-device continuity | Start a task on mobile, review on web, finish on desktop all synced |
Notably, OpenAI is sunsetting its standalone Atlas browser as part of this transition, folding its lessons into ChatGPT built-in browser instead a similar pattern to what happened with Operator.
What the cloud browser can’t do (yet)
Per OpenAI own documentation, the cloud browser component only works on public pages at launch it does not accept login credentials, use autofill, sign into websites, or complete payments. If a task requires any of those, it stops and hands back to you. This is a real limitation worth knowing if you expected an “it just logs in and buys things for me” experience.
Real Use Cases
OpenAI has published several customer accounts of ChatGPT Work in production:
Zapier used it to build a system tracing thousands of leads a month across CRM, email, and other tools, surfacing seven figures in previously-missed pipeline via a weekly dashboard.
RingCentral automated monthly product-launch checks across Jira and go to market schedules, letting one person effectively support roughly 50 product managers instead of one.
Virgin Atlantic used it to benchmark customer journeys against competitor airlines for a five year strategy plan, cutting a multi week analysis cycle down to hours.
NVIDIA automated pre and post-event prep for its GTC conference, cutting a process that previously ate roughly 40% of one employee’s pre-event time.
These are OpenAI published case studies, not independently verified third-party audits worth keeping in mind as you weigh them, though they do point to genuine enterprise-scale use.
Pricing (Verified Live, August 2026)
Prices accurate as of August 2, 2026. Verify current pricing before purchase, as OpenAI has changed plans multiple times in the past year.
| Plan | Price | Work Access |
|---|---|---|
| Free | $0/month | Limited, desktop app only |
| Go | $8/month | Not included |
| Plus | $20/month | Expanded web, mobile, and desktop |
| Pro | From $100/month | Maximum access, priority usage |
| Business / Enterprise / Edu | Custom | Full access, admin controls, spend limits |
Usage for ChatGPT Work follows the same credit based structure as Codex rather than a simple per message cap more complex, longer-running tasks consume more of your plan’s included usage.
Pros
1. Genuine capability upgrade over Operator: creates finished deliverables, not just browser actions
2. Deep integration with real business tools via 1,400+ connectors
3. Available to every paid plan tier, with even the Free desktop app getting limited access
4. Backed by ChatGPT Enterprise’s existing security, compliance, and admin-control foundation
5. Actively used internally at OpenAI across nearly all teams, per the company’s own account
Cons
1. Frequent renaming and re-architecting (three product identities in 18 months) makes it hard for casual users to know what’s current
2. Cloud browser can’t log into sites, use saved passwords, or complete purchases at launch a real gap versus the “does everything for you” pitch
3. No independent, third party benchmark exists yet to verify task completion performance
4. Usage based credit system is less predictable than a flat message allowance
5. Full rollout was staggered by plan tier, so access timing varied at launch
Limitations
[DATA UNAVAILABLE NOT PUBLISHED AS OF THIS WRITING]: No OSWorld, WebArena, or comparable third-party benchmark score has been published for ChatGPT Work. We are not estimating one from Operator old 2025 scores, since the underlying model and architecture have changed substantially. If Open AI or an independent lab publishes one, this article will be updated.
Comparison Table: Operator vs. ChatGPT agent vs. ChatGPT Work
| Operator | ChatGPT agent | ChatGPT Work | |
|---|---|---|---|
| Status | Discontinued (Aug 2025) | Discontinued (2026) | Active |
| Launch | Jan 2025 | Jul 2025 | Jul 2026 |
| Core model | CUA (GPT-4o based) | Unified agentic model | GPT-5.6 |
| Access | Pro only, U.S. only | Pro, Plus, Team | All paid plans + limited Free |
| Output | Browser actions only | Browser + research synthesis | Full documents, sheets, slides, sites |
| App connectors | None | Limited | 1,400+ |
| Scheduled/background tasks | No | Limited | Yes |
Final Verdict
Who is this for? Professionals and teams who want an AI system that can gather context from their actual tools and produce finished, usable work not just click around a browser.
Is it worth it? For Plus tier and above subscribers already inside the ChatGPT ecosystem, yes it’s a substantial capability upgrade delivered at no extra cost beyond your existing plan. For someone specifically hunting for “Operator” by name, the honest answer is that product doesn’t exist anymore, but what replaced it does more.
Bottom line: Open AI Operator was a real, useful first attempt at browser agents that lived for about seven months before being absorbed into something bigger twice. If you’re evaluating Open AI current agentic capabilities, evaluate ChatGPT Work, not Operator.
Who Should Use It
1. Teams that already pay for ChatGPT Plus or above and want to consolidate research, drafting, and light automation into one tool
2. Sales, finance, and marketing teams with existing SaaS tool stacks (Slack, Salesforce, Google Workspace) that ChatGPT Work can connect to
3. Individuals comfortable reviewing and approving AI-driven actions rather than expecting full autonomy
Who Should Avoid It
1. Anyone needing an agent to log into sites, use saved passwords, or complete purchases autonomously today the cloud browser doesn’t support this yet
2. Users on the Free or Go plans expecting full web/mobile Work access it’s desktop-only or unavailable at those tiers
3. Anyone who needs a verified, third-party benchmarked performance score before committing, since none exists publicly yet
Alternative Products
For readers comparing options in the AI agent space, other tools worth researching include Anthropic’s Claude (with its own agent and computer use capabilities), Google’s Gemini based agents, and Microsoft Copilot agent features each with different strengths, pricing, and enterprise integration depth. A dedicated comparison is a natural next read for anyone evaluating this category.
Future Outlook
Given OpenAI pace over the past 18 months three distinct product identities for essentially one evolving capability further changes to ChatGPT Work’s name, scope, or pricing within the next year would not be surprising. The clearest throughline across all three products is the underlying ambition: an AI that can act on your behalf across the web and your own tools, with increasing autonomy and increasing safeguards in parallel.
Frequently Asked Questions
Is OpenAI Operator still available? No. It was discontinued on August 31, 2025, after being folded into ChatGPT agent that July.
What replaced OpenAI Operator? Its capabilities live on inside ChatGPT Work, OpenAI current agent (launched July 9, 2026), following an intermediate step through “ChatGPT agent” (July 2025-2026).
Can I still use “agent mode” in ChatGPT? No per Open AI Help Centre, ChatGPT agent is no longer available. Use ChatGPT Work instead.
How much does ChatGPT Work cost? It requires a paid ChatGPT plan for full web/mobile access Plus ($20/month) and up though the desktop app offers limited access on every plan, including Free.
Does ChatGPT Work perform better than Operator did? No official apples to apples benchmark exists yet to confirm this quantitatively. Qualitatively, its feature set is substantially broader it creates finished work product rather than just performing browser actions.
Sources
ChatGPT agent – OpenAI Help Center
Introducing ChatGPT agent: bridging research and action – OpenAI
ChatGPT Work and Codex – OpenAI Help Center
Using cloud browser in ChatGPT – OpenAI Help Center
ChatGPT is now a partner for your most ambitious work – OpenAI
ChatGPT Work for every team – OpenAI
Don’t get caught off guard by the next AI pivot. OpenAI has renamed this product twice in 18 months and it won’t be the last time. Follow Bloxstation for the moment it happens next, not months later.
[Subscribe to Bloxstation →]
