ChatGPT Work vs Claude Cowork

OpenAI launched ChatGPT Work, its answer to Claude Cowork. If Cowork is the non-technical version of Claude Code, ChatGPT Work is the non-technical version of Codex. Same idea: an agent that works inside your apps, files, and folders instead of just chatting with you.

It launched alongside GPT-5.6, OpenAI's new model family, which powers the whole thing. Three tiers: Luna is the fastest, Terra balances speed and capability, and Sol is the most capable. You pick the model per task, the same way you pick between Haiku, Sonnet, and Opus.

Here's what it does, how it maps to Cowork, and whether you should care.

Where you can use it

Three places:

Web. Open ChatGPT and you'll see a Work toggle next to Chat at the top. Switch over and you get a task box, project workspaces, your connected apps, and a model selector.

Mobile. Same toggle in the app. Start a new chat, tap Chat at the top, switch to Work.

Desktop. The desktop app adds one thing the others don't: you can point it at specific folders on your computer, exactly like Cowork.

Tasks sync across devices. Start something on your laptop, check the output on your phone. Anthropic shipped the same sync for Cowork mobile around the same time, so both platforms now cover this.

The feature map

The two products are close to identical in structure. Different names, same concepts.

A "site" is what ChatGPT calls the rendered HTML output you'd know as an artifact. Ask for a morning briefing and it will pull from your calendar, inbox, and local weather, then hand you a formatted page you can open and use. Same experience as Cowork, different label.

Your skills are portable

This is the part that matters most if you've invested time in Cowork.

Skills in both platforms are just SKILL.md files: plain instructions that tell the agent how to do a specific task. If you've built up a library of them in Claude, they are not locked in. Download the skill file from Claude, then in ChatGPT Work go to Plugins, click Skills, click Add, and upload the file. Done. Run it by typing @ and the skill name.

The same instructions produce the same kind of output on either platform. One difference worth knowing: ChatGPT has native image generation. If your skill produces images, it runs without needing an external image tool connected. In Claude you'd wire that up through a connector.

Connecting your apps

Plugins cover the usual suspects: Gmail, Google Calendar, Slack, Drive, and so on. Browse the plugin directory, connect what you use, and the agent can read from and act in those apps.

If an app you rely on isn't in the directory, the Zapier MCP fills the gap. Zapier connects to 9,000+ apps and exposes them as an MCP server you can add manually. Create a new MCP server in Zapier, choose ChatGPT, enable developer mode in ChatGPT, paste in the server URL, sign in to Zapier, then add whatever apps you need. Beehiiv, for example, has no direct ChatGPT plugin, but through Zapier it connects in seconds. The same trick works in Claude.

Scheduled tasks

Both platforms run recurring automations: daily email briefs, inbox monitoring, content trend scans, whatever you've set up. Anything scheduled in Cowork can be rebuilt in ChatGPT Work because the underlying pieces (a skill plus a schedule plus connected apps) are the same.

Should you switch?

No. Not because ChatGPT Work is bad. Because the two are so similar that switching buys you almost nothing.

The real takeaway is that your setup is no longer tied to one platform. Skills are markdown files. Connectors exist on both sides. Scheduled tasks rebuild in minutes. When OpenAI ships a model that beats the current Anthropic one, you can test your exact workflow over there. When Anthropic ships the next Claude, you can move back. You're building an operating system, and the platform underneath it is now swappable.

If you're already productive in Cowork, stay. If you're starting fresh, pick whichever ecosystem you already pay for. Either way, keep your skills in files you control. That's the actual moat.