ChatGPT Now Has Chat and Work, So Where Does Office Work Go?

If you've opened ChatGPT lately, there are two options at the top: Chat and Work. On teams, the question that keeps coming up is simple but confusing: "Where should this go?" I use both every day, and after a few months the pattern is fairly clear.

The difference in one sentence

According to OpenAI's help page on Work and Codex, Chat is for fast, conversational help and everyday questions, while Work is an agent for longer, multi-step work that ends in finished deliverables. Codex stays dedicated to software work.

Work itself was introduced on July 9, 2026. In the ChatGPT release notes, OpenAI says it can research and analyze information, work across connected apps and files, and create documents, spreadsheets, presentations, reports and Sites. You can follow its progress, answer its questions, change direction and approve important actions while it works.

Where office work goes

This is the split I use, put together from OpenAI's guidance and my own experience:

The easy rule: if you need an answer, use Chat. If you need finished work, use Work.

Examples from everyday work

• A monthly sales recap from several spreadsheets goes to Work. It can read the files, do the math, then write a summary or a deck. If the report is routine, Work can run it on a schedule. I covered when a monthly report is really worth automating in Your Team Stays Up Every Month-End Compiling Reports: When Is It Worth Automating?
• Preparing a quotation on the go also goes to Work, because the result is a document. My story about that is in I Planned to Stop at a Café to Work, but the Quotation Was Done While I Talked to ChatGPT.
• Answering a customer question or tidying up a paragraph is fine in Chat. Opening Work for this only takes longer.
• Looking up a rule or a term also belongs in Chat, then check the source.

The models are different too

OpenAI's help page lists GPT-6.1 Sol, GPT-6 Sol and GPT-6 Luna as the models for Work and Codex. Meanwhile, since October 7, 2026, Chat runs GPT-6 with Intelligent UI, which can mix text, visuals and interactive elements like calculators into an answer. According to the same release notes, the models powering Work and Codex didn't change with that update.

For everyday users, there's no need to memorize the models. What matters is that Chat and Work are built for different kinds of work, so it's normal for their results and speed to differ.

What to watch for

1. Usage is separate. Work follows the same usage structure as Codex, and how much it uses depends on the task. Heavy tasks use up the allowance faster, so don't open Work for a question Chat can answer.
2. Not every plan gets Work. At launch, Work was available on paid plans except Free and Go.
3. In a company workspace, admins decide. Workspace owners and admins can control Work access by role. Chat stays available in every configuration.
4. Important actions still need your approval. If Work wants to send an email or change something in a connected app, you approve it. It's the same principle I discussed in At What Point Should Humans Still Check AI and Automation Results.
5. The data you bring is still your responsibility. Work can read the files and apps you connect, so think first about which data is safe to bring, as I covered in Want to Summarize Your Company's Financial Data in ChatGPT? Is It Dangerous?

Closing

Chat and Work aren't about which one is smarter, but about the kind of work. Quick questions go to Chat, work that ends in a finished deliverable goes to Work. Once a team agrees on that split, usage goes further and the results are more dependable.

If you want to figure out which tasks on your team can go to AI and which are better automated with your own system, tell me what you need.

The photos in this article were made with AI, not photos of real people or places.