Your colleague drops a task on Friday afternoon - "pull the Q2 churn data, build a deck, send the draft to the client folder." On Monday morning it's done. That is the promise OpenAI made on July 9, 2026 when it launched ChatGPT Work, three months after Anthropic's Claude Cowork reached general availability. Two frontier labs, same thesis, shipped inside one quarter.
The interesting question is not whether these are real. They are. On July 9, 2026, OpenAI announced ChatGPT Work - a work agent that researches materials, operates files and apps, and completes slides, spreadsheets, and documents.
Anthropic had been selling the same promise since January: Claude Cowork went GA in April and expanded to mobile and web two days before OpenAI's launch event. The interesting question is where each one actually breaks, and what that means for a team that wants to hand one a real job.
What each one actually does
Claude Cowork reached general availability on April 9, 2026, and ChatGPT Work launched July 9. Both let you describe an outcome instead of writing a prompt, then work independently for hours and hand back finished files. At that level they look identical. The differences show up in environment, scheduling, and how much of your stack each one can reach.
Comparing them is really a choice between "an agent that lives inside your desktop and touches your files" and "an agent that lives in your browser and touches the web." Concretely: Cowork excels at local file pipelines - analyze Excel, build a presentation, draft an email. Work excels at web-based workflows - research, booking, data gathering from multiple sites.
Scheduling is the sharpest practical difference. Claude Cowork has Tasks, and a Task can be set to run on a cadence - every morning, every Monday, first of the month. Since the July update moved web and mobile execution onto Anthropic's servers, those scheduled runs fire with none of your devices online.
ChatGPT Work starts each task and it finishes when the work is done. There is no recurring trigger.
| Claude Cowork | ChatGPT Work | |
|---|---|---|
| GA date | April 9, 2026 | July 9, 2026 |
| Primary environment | Desktop, local files | Browser, cloud SaaS |
| Scheduled recurring tasks | Yes (server-side since July) | No |
| Microsoft 365 depth | Read + write (July update) | Limited at launch |
| App connector count | Curated MCP ecosystem | 1,400+ apps |
| Underlying model | Claude Opus 4.8 | GPT-5.6 Sol |
| Per-task pricing | No - draws on plan allowance | No - draws on plan allowance |
ChatGPT Work has the larger connector directory - 1,400+ apps - and ships polished artifacts like interactive web apps; Claude Cowork has months more production maturity, desktop-level access to local files, and Anthropic's vetted MCP connector ecosystem.
On Microsoft 365 specifically, Cowork currently leads. Claude Cowork has had the M365 MCP Connector since February 2026; in July 2026, Anthropic added write capabilities - Claude can now draft and send emails, manage calendar events, and create or update files in OneDrive and SharePoint. ChatGPT Work does not yet have equivalent Microsoft 365 depth.
The compliance gap nobody flagged at launch
This is the thing worth knowing before you send either agent to touch a business document. As of July 10, 2026, Anthropic's official help explicitly states that 'Cowork activity is not currently captured by the Compliance API.' That is a meaningful hole for any regulated team. If your legal or security team relies on audit logs to review AI-generated content before it reaches clients or gets filed, Cowork's outputs are currently invisible to those logs. You can still use it - but you are running on the honor system until Anthropic closes the gap.
ChatGPT Work inherits OpenAI's enterprise audit infrastructure, which is more mature. The two products are built on different models and make different tradeoffs on autonomy versus safety. Choosing between them is less about which is "better" and more about which fits your organization's existing stack and risk tolerance.
Enterprise admins get spend controls; individuals find out empirically how fast an hours-long agent burns a monthly allowance. Run a test task and watch the usage meter before you automate anything weekly.
Beagle in action#ops, Monday 9:02amWhere Cowork has the edge, and where Work does
The Composio Golden Eval benchmark ran 47 business scenarios across both agents. The eval contained 47 scenarios and 94 trials - one run per scenario with Claude Cowork and Claude Fable 5, and one with ChatGPT Work and GPT-5.6 Sol at high reasoning. Both used Composio MCP to access third-party SaaS apps including HubSpot, PostHog, and Zendesk.
The headline from that benchmark: ChatGPT Work on GPT-5.6 is better at generating polished deliverables - presentations, formatted reports. Claude Cowork on Opus 4.8 is better at analytical tasks requiring deep reasoning and accuracy.
Claude Cowork is better for inline visuals, Artifacts, and dashboards connected to app data. ChatGPT Work is better when the output needs to become a hosted site with a shareable URL.
Two things are genuinely non-obvious here. First, the product lineups now map onto each other almost exactly - OpenAI has Chat, Work, and Codex; Anthropic has Claude, Cowork, and Claude Code. When two labs independently arrive at identical three-product structures within months of each other, that shape is probably right. Second, the fact that both products lack per-task billing is a deliberate choice, and it has consequences. An agent working for six hours on a complex task consumes the same plan resource as 200 quick chat turns. Teams that run frequent long tasks will hit their ceiling faster than their current usage patterns predict.
The approval loop is the point. Approval-gated execution and mid-task steering matter. Cowork lets you interrupt a running task and redirect it without starting over. ChatGPT Work runs to completion and hands back the result. Neither model is better by default - it depends on whether you want to steer mid-flight or just review the landing.
ChatGPT Work vs Claude Cowork: common questions
Which is better for teams that live in Microsoft 365?
Claude Cowork currently has the stronger Microsoft 365 story. It added write access to Outlook, OneDrive, SharePoint, and Teams in July 2026, meaning it can draft and send emails and create files - not just read them. ChatGPT Work has announced broader integration plans but had not shipped equivalent depth at launch.
Do these agents charge per task?
Neither charges per task. Both bill against your existing plan allowance - the same pool you draw from for regular chat. A long agentic run can consume several hours of equivalent usage. Enterprise admins can set spend controls; individual plan holders have less visibility until the allowance runs low.
Can I use ChatGPT Work or Claude Cowork in Slack?
Neither integrates natively into Slack as a first-party trigger. You can start a task from web or mobile, but the agent runs in its own environment and returns output there. A teammate like Beagle can bridge the gap - picking up a request in Slack, handing context to an agent, and posting the result back in-thread for approval. See Beagle's use cases for how that handoff works.
Is Cowork output logged for enterprise compliance?
Not as of July 10, 2026. Anthropic's own help documentation states that Cowork activity is not currently captured by the Compliance API. If your team relies on audit logs for AI-generated content, treat Cowork output as unlogged until that changes. ChatGPT Work sits inside OpenAI's enterprise audit infrastructure, which is more established.
Which agent handles scheduled, recurring tasks?
Claude Cowork does, and since July 2026 those scheduled tasks run on Anthropic's servers - not your local machine. ChatGPT Work, as of its July 9 launch, does not support recurring task scheduling. If you need an agent to run a weekly report without a human trigger, Cowork is the current choice.