ChatGPT Work (2026): Pricing, Limits & Alternatives
OpenAI launched ChatGPT Work on July 9, 2026, an agent that turns a goal into a finished report, spreadsheet, or site. See pricing, limits, and rivals.
ChatGPT Work replaces the manual routine of pulling data from several apps to finish one deliverable. It suits operations, sales, and marketing teams who need a report, spreadsheet, or site built from a brief rather than typed line by line. It runs unattended for hours on end and checks in only at set approval points, not fully hands-off.
ChatGPT Work is OpenAI's agentic mode inside ChatGPT, launched July 9, 2026 and powered by the GPT-5.6 model family. It plans a stated goal into steps, pulls context from connected business apps and files, and finishes documents, spreadsheets, presentations, or websites without constant prompting. It pauses only for defined approval checkpoints.
Maker: OpenAI · Autonomy: semi autonomous · Maturity: GA
Underlying models: GPT-5.6 Sol, GPT-5.6 Terra, GPT-5.6 Luna
About ChatGPT Work
ChatGPT Work is OpenAI's workplace agent mode inside ChatGPT, launched July 9, 2026. It turns a plain-language goal into a finished document, spreadsheet, presentation, or website by pulling context from a user's connected apps and files, instead of just answering questions in a chat window. OpenAI built it to close the gap between a chatbot that talks about work and an agent that actually produces the deliverable. Work runs on GPT-5.6 Sol, OpenAI's flagship model in the GPT-5.6 family, with Terra and Luna covering everyday and cost-efficient tasks. It plans a project into steps, pulls information from connected tools through "@" plugin commands (Slack, Microsoft Teams, Gmail, Google Drive, Salesforce, SharePoint), browses the web with a built-in browser, and can keep working on a task in the background for hours. Human Checkpoint Triggers pause the agent for approval before high-stakes actions like sending an external message or touching a production system, so it behaves as a semi-autonomous agent rather than a fully hands-off one. It suits people who currently stitch together several tools to finish one deliverable: operations managers building recurring reports, revenue teams updating CRM records from meeting notes, and marketing teams turning a brief into a shareable site through the Sites beta. Scheduled Tasks let a team automate recurring work, such as refreshing a presentation whenever feedback arrives by email. Work is available on ChatGPT Business and ChatGPT Enterprise seats, not on Free, Go, or standalone Plus. Enterprise buyers negotiate pricing directly with OpenAI, while Business is self-serve with online checkout. Access rolled out first to Pro, Pro Lite, Enterprise, and Edu accounts before reaching remaining paid tiers, and Work runs on web and the merged desktop app for macOS and Windows, with Codex Remote support on mobile. OpenAI has shipped updates on a roughly weekly cadence since launch: a merged desktop app for Chat, Work, and Codex, Codex Remote reaching general availability with a DigitalOcean Droplet Workspace plugin, and voice control for starting and monitoring Work tasks from a phone.
Pricing
ChatGPT Work ships inside ChatGPT Business ($20/seat/month billed annually, $25 monthly, 2-seat minimum) and ChatGPT Enterprise (custom quote, reported $45 to $75/seat/month, roughly 150-seat minimum, annual prepay). Not available on Free, Go, or standalone Plus plans; heavier usage now draws on a credit-based system introduced after the introductory period.
Key Features
- Goal to finished deliverable: Breaks one plain-language prompt into a working plan, then executes each step itself and stays on a single project in the background for hours at a stretch.
- Native app connectors: Pulls live context from Slack, Microsoft Teams, Gmail, Google Drive, Salesforce, and SharePoint through "@" plugin commands instead of manual copy-paste.
- Human Checkpoint Triggers: Pauses for explicit approval before high-stakes actions such as sending external messages or editing production data, keeping a human in the loop on risk.
- Built-in browser and Computer Use: Browses the live web and can click, type, and move files across desktop apps to finish tasks that span multiple programs.
- Scheduled Tasks: Automates recurring work like refreshing a report when new data lands or updating a meeting agenda after a Slack thread changes.
- Sites: Turns a plan or dataset into a shareable interactive website through the Sites beta, without a separate web builder.
Strengths
- Bundled directly into existing Business and Enterprise ChatGPT seats instead of sold as a separate SKU, so teams already on a paid plan get it with no new contract.
- Natively connects to 10+ business apps, Slack, Salesforce, SharePoint, Google Drive, and Gmail among them, through "@" commands that remove manual copy-paste between tools.
- Keeps running on a single goal for hours unattended, checking back in only at defined approval checkpoints instead of needing a prompt per step.
Weaknesses
- Enterprise pricing is quote-only and requires talking to sales, putting Enterprise-only compliance features like the HIPAA BAA and data residency out of easy reach for small teams that just want to try them.
- Blocked entirely in the EEA and UK, and unavailable on the Free and Go plans.
- OpenAI has not published a Work-specific SWE-bench or GAIA score, so buyers comparing agentic coding quality against Claude Code have to rely on the underlying model's general benchmarks instead of a Work-specific number.
Frequently Asked Questions
How much does ChatGPT Work cost in 2026?
ChatGPT Work is bundled into ChatGPT Business, priced at $20 per seat per month billed annually or $25 billed monthly with a 2-seat minimum, and ChatGPT Enterprise, which is quote-only and reported to run $45 to $75 per seat per month with a roughly 150-seat minimum and annual prepay. There is no separate charge for Work itself; usage beyond plan limits moves to credit-based pricing after the free trial period ends.
Is ChatGPT Work free to use?
ChatGPT Work has no free tier: it is not available on the Free or Go plans. OpenAI offered a limited introductory trial for workspace agents at launch, after which every task requires a paid Business or Enterprise seat.
What are the best alternatives to ChatGPT Work?
Claude Cowork runs as a sandboxed desktop agent on Claude Opus 4.6 and is a closer fit if document creation quality matters more than app connectors. MultiOn focuses narrowly on browser automation rather than full workspace tasks, and is lighter weight if clicking through web forms is the whole job. Eigent offers an open-source, self-hosted alternative for teams that want the agent running locally instead of in OpenAI's cloud.
How does ChatGPT Work compare to Claude Cowork in 2026?
ChatGPT Work leans on native business-app connectors like Slack and Salesforce, while Claude Cowork runs in a sandboxed Linux VM on Claude Opus 4.6 and focuses on producing polished files such as .docx, .xlsx, and .pptx. On raw coding benchmarks, Claude's models still lead: Claude Opus 4.8 scores 88.6% on SWE-bench Verified versus roughly 88.7% for GPT-5.5, essentially tied, with Claude Mythos Preview ahead at 93.9%. Teams centered on Microsoft and Google business apps tend to prefer Work; teams centered on document polish tend to prefer Cowork.
How do you get started with ChatGPT Work?
Sign in to a ChatGPT Business or Enterprise account, switch to the Work tab in the desktop or web app, and connect the business apps you need through the connectors panel. Give Work a plain-language goal, such as building a weekly sales report from CRM data, and it plans the steps and starts pulling context immediately, checking in only when it hits an approval checkpoint.
Top Alternatives
- MultiOn: Pick MultiOn if you only need a browser agent clicking through web forms; pick ChatGPT Work if you need it wired into Slack, Salesforce, and SharePoint too.
- Eigent: Pick Eigent if you want an open-source multi-agent desktop app running locally; pick ChatGPT Work if you want a managed cloud agent bundled into an existing ChatGPT seat.
- Claude Code: Pick Claude Code if the whole job is agentic coding in a terminal; pick ChatGPT Work if the job spans documents, spreadsheets, and code together.