OpenAI Launches ChatGPT Work + GPT-5.6 — and Starts Sunsetting the Atlas Browser (July 2026)

News, July 2026. On July 9, OpenAI launched ChatGPT Work, an agent built to carry out entire jobs rather than answer questions — and released the model behind it, GPT-5.6. It’s a concrete step past the chat box: you hand it an outcome, and it gathers what it needs across your connected apps and the web, then ships the finished artifact. For anyone thinking about AI visibility, the interesting part isn’t the demo — it’s that a new surface just appeared where your content gets used without a person ever loading your page.

What OpenAI Shipped

The headline product is ChatGPT Work. Where the assistant answered a prompt, the agent takes an outcome — “build this report,” “put together this deck” — then gathers information across your connected apps and workflows, breaks the job into smaller steps, and completes them on its own. OpenAI says it can stay with complex projects for hours and keep working while you’re away from your computer or phone, producing documents, spreadsheets, presentations, and even web applications. It started rolling out on web and mobile to Pro, Enterprise, and Edu users, expanding to Plus and Business over the following days.

Underneath it is GPT-5.6, released the same day in three tiers: Sol (top), Terra (mid-range), and Luna (fast and cheap). The release also folds Codex into a single upgraded ChatGPT desktop app and adds a hosted sites service for OpenAI customers. Taken together, it’s less a new chatbot and more an attempt to make ChatGPT the place work gets done, not just asked about.

The Atlas Browser Sunset

The quieter, more telling detail: OpenAI is beginning to sunset the standalone Atlas browser it launched in October 2025. The agentic-browser bet — a dedicated browser that navigates the web for you — is being folded into the agent itself. Rather than a separate app that browses, the browsing capability now lives inside ChatGPT Work, which reaches across the web and your apps as part of completing a job.

That’s a meaningful reframing of how these systems touch the web. The unit is no longer “a browser session a user drives” but “an agent that pulls what it needs to finish a task.” The web-facing behavior — fetch, read, extract, use — is the same; what changed is that it’s now a step inside a workflow rather than a destination a person visits.

What It Means for AI Visibility

For AI visibility, the practical takeaway is continuity, not a new rulebook. A work-agent that assembles a deliverable still has to choose sources it can find, extract cleanly, and trust — the same selection pressure that governs AI answers, now applied to producing a document instead of a reply. If your content is extractable, front-loads the specifics, and comes from an entity the model can resolve, it’s a candidate for the agent’s output; if it isn’t, the agent reaches for a competitor that is.

This is the surface our AI agent SEO guide is about — optimizing for agents that browse and do work rather than users who read — and ChatGPT Work is the clearest instance of it yet, with the Atlas sunset showing the agentic-browser idea collapsing into the agent. The move to make: keep being the source an agent can safely pull from, and watch whether the engines you care about actually cite you.

FAQ