The Short Version
OpenAI released ChatGPT Work in July 2026 — a new mode inside ChatGPT that runs longer, more complex tasks autonomously, working across connected apps, local files, and browsers rather than just answering questions in a chat window.
It's built on GPT-5.6, OpenAI's latest frontier model. It ships with Scheduled Tasks (recurring work that runs on its own), Computer Use (the agent can click and navigate apps on your desktop), and Sites (currently in public beta — a way to publish what ChatGPT builds as a live interactive page).
On desktop, ChatGPT Work is available on every plan including Free. On web and mobile, the initial rollout is for Pro, Enterprise, and Edu users, expanding to Plus and Business over the following days.
The biggest practical difference from regular ChatGPT: this is designed for tasks that take hours, not seconds. That changes how usage is counted — and for Enterprise and Edu admins specifically, how it's managed.
What ChatGPT Work Actually Is
Up until now, ChatGPT worked in turns: you send a message, it responds, done. That interaction model works fine for drafting emails or summarizing documents. It doesn't work well for anything that requires sequencing steps, waiting for external inputs, looping back when something changes, or running in the background while you do other things.
ChatGPT Work is OpenAI's answer to that gap. According to OpenAI's announcement, it's designed to "take on more ambitious tasks" — specifically ones that involve gathering information across multiple apps, creating finished materials like slides, sheets, docs, and web apps, and staying with complex projects for hours by breaking them into smaller steps and completing them independently.
The word "independently" matters here. This isn't a smarter autocomplete. It's an agent that can hold a multi-step plan, use tools to execute pieces of it, check its own work, and hand the result back when it's done.
That's a meaningful shift in how ChatGPT fits into a business workflow — and it's worth understanding what the system can actually do before making decisions about adopting it.
The Codex Foundation
To understand why ChatGPT Work works the way it does, it helps to understand where it came from.
Codex was OpenAI's coding agent — originally built for developers to write, debug, and ship code with less manual intervention. It was architected from the start to run multi-step tasks, use tools, and operate autonomously over longer time horizons.
OpenAI has now merged Codex into the main ChatGPT desktop app. Starting with this release, the ChatGPT desktop app includes Chat, Work, and Codex as three modes within a single application. The Codex app, which existed separately, is becoming the new ChatGPT desktop app — existing Codex users can update as usual, and desktop Codex projects remain accessible through the ChatGPT mobile app.
ChatGPT Work inherits Codex's agentic architecture. That's what enables the longer-running, multi-step behavior. OpenAI notes that more than 5 million people use Codex every week, and more than 1 million now use it for work outside software development — suggesting the capability had already found a broader audience before this release.
What It Can Actually Do
Here's what's shipping, based on OpenAI's announcement:
Connected apps via plugins
ChatGPT Work uses plugins to connect to tools where work already happens: Slack, Microsoft Teams, Google Drive, SharePoint, email, calendars, CRMs, and project trackers. Once connected, you can direct it to pull context from a specific app by typing "@" followed by the app name in your prompt — or let it decide automatically which sources are relevant.
Document and output creation
From those connected sources, ChatGPT Work can produce slides, sheets, docs, analyses, and web content. The core use case is turning raw inputs — discovery call notes, source data, customer research — into finished or near-finished materials that your team can use or review.
Computer Use (desktop only)
On the desktop app, ChatGPT Work can use your computer directly: clicking, typing, navigating apps, and moving files. OpenAI describes this as letting ChatGPT "execute tasks in the background across your apps, tools, and browser." This enables tasks that require interacting with local software that doesn't have a plugin or API.
Scheduled Tasks
ChatGPT Work can run on a schedule or in response to events — not just when you prompt it. Examples from OpenAI's announcement include reviewing new Slack updates each week and refreshing a recurring meeting agenda, checking websites and dashboards each morning and summarizing what changed, or monitoring for new customer feedback and turning recurring themes into prioritized product ideas.
This is the capability that makes ChatGPT Work genuinely different from a chat interface: you can set something up once and have it run without you.
Sites (public beta)
OpenAI is also introducing Sites — a way to turn work or ideas into an interactive site or web app, shareable via URL. The announcement describes use cases like live dashboards, project trackers, launch calendars, prototypes, and interactive reports. Sites is currently in public beta. OpenAI says ChatGPT can also update Sites as the underlying information changes.
Built-in browser (desktop)
The desktop app now includes a built-in browser that lets ChatGPT research markets, pull information from websites, compare sources, and open files from Google Workspace and Microsoft 365 inside the app.
The GPT-5.6 Model Underneath
ChatGPT Work runs on GPT-5.6, OpenAI's new frontier model family released alongside this announcement. The family spans three tiers: Sol, the flagship; Terra, a balanced model for everyday work; and Luna, the most cost-efficient option.
Which tier you get in ChatGPT Work depends on your plan:
- Free and Go users access GPT-5.6 Terra.
- Plus, Pro, Business, and Enterprise users can choose among Sol, Terra, and Luna, and set an effort level for each.
max— a higher-reasoning setting — is available to all users who have GPT-5.6 access in ChatGPT Work and Codex, and can be toggled on in settings.ultra— OpenAI's highest-capability setting, which coordinates multiple agents across parallel workstreams for the most demanding tasks — is available in ChatGPT Work to Pro and Enterprise users.
The reason the model matters: multi-step agentic work requires a model that can reason through intermediate steps without losing track of the original goal, handle ambiguity in instructions, and produce outputs that match provided templates. OpenAI says GPT-5.6 is designed specifically for this — "state of the art at reasoning through multi-step tasks and creating materials that follow your templates and reference files."
How Usage Works — and Why It's Different
ChatGPT Work's usage model is worth reading carefully before rolling it out to a team.
OpenAI states this directly in the announcement: "ChatGPT Work is designed for longer, more involved work than a typical chat request, so usage works differently. Usage varies with the amount of work required, and more complex tasks may use more of your plan's included usage."
What this means: your plan includes a usage allowance. A quick chat message uses a small portion of it. A ChatGPT Work task that runs for hours, makes dozens of tool calls, and produces a multi-part output uses proportionally more. The allowance doesn't change; what changes is how quickly a complex task draws it down compared to a simple one.
This is not a new pricing tier. OpenAI announced no price increase for ChatGPT plans with this release. It's a usage accounting reality — agentic work is computationally heavier than a single-turn conversation, and that's reflected in how your existing allowance is consumed.
For most individual users on Plus or Pro, this is unlikely to matter day-to-day unless you're running large batches of long-running tasks. For teams rolling out ChatGPT Work at scale, it's worth factoring into adoption planning.
Admin Controls for Enterprise and Edu
For Enterprise and Edu organizations, OpenAI is providing dedicated tools to manage ChatGPT Work usage as it spreads across teams.
Per the announcement, Enterprise and Edu admins can set spend controls in the Admin Console to manage ChatGPT Work usage. The controls include:
- Setting workspace-level defaults
- Configuring group-level limits
- Creating individual overrides for people who need more capacity
- Reviewing requests for additional credits with user-submitted project details and rationale
This is an admin-level capability. It is not available to individual Plus or Pro users — those plans don't have an Admin Console or organizational usage management.
The framing in the announcement is worth noting: "Admins can support high-impact work across teams without raising limits broadly." This is a governance tool, not a cost-cutting one. The intended use is precision — making sure capacity goes to the work that justifies it, rather than either restricting everyone or giving everyone unlimited access.
There's also an Auto-review feature: a separate model reviews important actions involving connected tools and APIs before they happen, helping prevent unauthorized sharing of sensitive information. For organizations handling client data or proprietary information through connected CRMs and document systems, this is the control that matters most.
What's Available to Whom — A Clean Map
This is where coverage has gotten muddled, so here's the breakdown as stated in OpenAI's announcements:
ChatGPT desktop app (macOS and Windows)
- Chat, Work, and Codex: available on every plan, including Free
- Available globally, download now
ChatGPT Work on web and mobile
- Initial rollout: Pro, Enterprise, and Edu
- Rolling out over the following days: Plus and Business
GPT-5.6 models inside ChatGPT Work
- Free and Go: GPT-5.6 Terra
- Plus, Pro, Business, Enterprise: choice of Sol, Terra, or Luna, with effort level
ultra: Pro and Enterprise
Sites
- Public beta
Computer Use
- Desktop only; part of the desktop app available on all plans
Admin spend controls
- Enterprise and Edu admins only, via Admin Console
The desktop app distinction is meaningful: if you're on a Free plan and want to try ChatGPT Work, downloading the desktop app gives you access today, running GPT-5.6 Terra.
What This Means Practically
A few observations for business owners and teams thinking about how this fits into their stack:
The "agent" framing is accurate, not hype. ChatGPT Work can run for hours without human input, use tools, and produce finished outputs. That's substantively different from what ChatGPT was before. The capability is real.
Plugins are the integration layer. The value of ChatGPT Work scales directly with what you connect it to. A standalone ChatGPT Work session without plugins is a capable but isolated tool. Connected to your CRM, project tracker, Slack, and Google Drive, it becomes a coordinator that can synthesize across those systems. The plugin setup step isn't optional if you want the full capability.
Scheduled Tasks may be the highest-leverage feature for most teams. One-time tasks are useful. Repeating tasks that update automatically — refreshed briefings, monitored dashboards, weekly summaries — are where persistent value compounds. This is the feature worth experimenting with first for teams with recurring information workflows.
Sites is early. It's in public beta. That means the feature set will change, stability is not guaranteed, and it's not ready for production use cases that require reliability. It's worth watching. It's not worth building around yet.
Enterprise admins need a deployment plan. Because complex tasks draw down usage allowances proportionally, deploying ChatGPT Work broadly without governance could produce uneven results — some teams use it heavily, others don't, and allowances get consumed unevenly. The spend controls exist to solve this, but they require intentional setup, not just activation.
The Bigger Picture
It's worth stepping back from the feature list to think about what this release represents structurally.
Codex was always the underlying technology. Building it into the main ChatGPT product — on every plan, on desktop, accessible without a separate app — lowers the adoption barrier significantly. OpenAI isn't positioning ChatGPT Work as a developer tool. They're positioning it as infrastructure for every knowledge worker.
The use cases in the announcement are explicitly cross-functional: sales teams preparing proof-of-concept materials, finance teams accelerating month-end close, marketing teams turning research into campaign assets. These are not technical users. The pitch is that agentic AI is now usable by anyone who can write a prompt and connect a plugin.
Whether that's true at scale — whether ChatGPT Work performs as reliably in a real organization as it does in controlled demos — is a question that will get answered over the next several months of adoption. The capabilities announced are real. The reliability and consistency at production scale is what's being proven now.
For a team that's been using ChatGPT primarily as a chat interface, this release is worth revisiting. The tool has changed. What you can reasonably ask it to do has changed. The question worth exploring is: which recurring, multi-step work in your organization is currently done manually because no tool could hold the full context and execute the steps? That's the use case ChatGPT Work was built for.
