OpenAI just rebuilt ChatGPT from the ground up (here's what actually changed)

If you blinked this week, you missed three separate product launches from OpenAI. Not three features. Three launches — a new frontier model family, a rebuilt ChatGPT app with an agent that does actual work for you, and a voice mode that finally stops interrupting you mid-sentence. Most companies space these out over a quarter. OpenAI dropped them within days of each other.

That's not an accident. It's a signal that OpenAI is done treating ChatGPT as a chatbot and is repositioning it as a work platform that competes directly with hiring a junior analyst — or at least a very capable intern who never sleeps.

Here's what actually shipped, why it matters, and how to start using it without wasting a week figuring out where everything moved.

What is GPT-5.6 and why does it have three names now?

OpenAI's GPT-5.6 announcement introduces a new naming system that ditches the old "just add a number" approach. The number identifies a model's generation, while Sol, Terra, and Luna identify durable capability tiers that can advance on their own cadence.

In plain terms:

Sol is the flagship — OpenAI says that GPT–5.6 Sol achieves "state-of-the-art results" in coding, knowledge work, cybersecurity, and science.
Terra is the everyday workhorse, built to balance power and speed for everyday work.
Luna is the budget option, OpenAI's most cost-effective model in the family.

The benchmarks are genuinely impressive. GPT-5.6 Sol comes within one point of Anthropic's Claude Fable 5 on the Artificial Analysis Intelligence Index (59 vs 59.9) while costing roughly one-third as much per task, and it leads the Coding Agent Index at 80 points. On pricing, Sol runs $5/$30 per million input/output tokens, Terra runs $2.50/$15, and Luna runs $1/$6 — meaningfully cheaper than comparable offerings from competitors.

There's also a new reasoning layer worth knowing about. Max mode allocates additional compute for complex problems, while Ultra coordinates four agents in parallel for demanding workflows. If you've ever wished ChatGPT would just spend more time thinking before answering, Max and Ultra modes are the dial for that.

What is ChatGPT Work and how is it different from regular ChatGPT?

This is the part that actually changes how you use ChatGPT day to day. OpenAI's official ChatGPT Work page describes it as an agent that gathers context, plans the approach, and takes action across your tools, files, and desktop apps to create polished spreadsheets, docs, and slides.

The core idea: instead of you copy-pasting information into ChatGPT and asking it to summarize, you point ChatGPT Work at your actual workflow and it goes and gets the information itself. It's an agent in ChatGPT that helps you take on more ambitious tasks, gathering information across your apps and workflows to create finished materials like sheets, slides, docs, and web apps. And it doesn't need constant supervision — it can stay with complex projects for hours by breaking them into smaller steps and completing them independently.

Under the hood, this is powered by Codex — yes, the same Codex that used to be a separate coding-only app. With Codex technology built in, ChatGPT can now move beyond answering questions to getting real work done across web, mobile, and desktop, and more than 5 million people already use Codex every week, with over 1 million using it for work outside software development. That last stat is the tell — people were already using a coding tool for non-coding tasks, so OpenAI just built that behavior directly into the main app.

According to OpenAI's help center documentation on ChatGPT Work, the app now has three distinct modes: Chat is for fast, conversational assistance and everyday questions, Work is an agent designed for longer, multi-step work and finished deliverables, and Codex remains dedicated to software development and technical work.

What can you actually connect it to?

This is where it gets genuinely useful for office work. You can connect plugins for Slack, Microsoft Teams, Google Drive, SharePoint, email, calendars, CRMs, project trackers, and other internal systems. Once connected, Scheduled Tasks let it run on autopilot — ChatGPT Work can perform one time or recurring actions, such as refreshing meeting agendas from Slack updates, checking websites and dashboards each morning, summarizing changes, monitoring customer feedback, or updating a presentation when new email feedback arrives.

On desktop specifically, it gets even more hands-on. ChatGPT on desktop can use local files and apps, while a built-in browser lets it gather information from websites, online tools, and cloud files. And in a move that mirrors what Anthropic has been doing with Claude's computer use features, Computer Use lets ChatGPT use your computer on your behalf to execute tasks in the background across your apps, tools, and browser — clicking, typing, and moving files where they need to go.

If you're worried about an AI agent having that much access, OpenAI has built in admin controls specifically for this. Admins can manage access to plugins, connected tools, browser use, network access, local desktop capabilities, and sensitive actions, and OpenAI said auto review blocked 100% of red team attempts to extract protected data during adversarial testing. Worth verifying independently before you connect anything sensitive, but at least it's not an afterthought.

How do you get access?

Rollout is staggered by plan. ChatGPT Work rolled out first for those on ChatGPT Pro, Enterprise, and Edu plans, with Plus and Business following shortly after. If you're not seeing it yet, that's likely why — ChatGPT Work is gradually rolling out to eligible accounts over the coming days, and if you don't see it yet, your account may not have access, but availability will continue to expand as the rollout progresses.

What happened to the old Codex app?

It's being absorbed. The standalone Codex app is merging into the new ChatGPT desktop app, while the older ChatGPT desktop app will be renamed ChatGPT Classic. So if your team built workflows around the standalone Codex desktop app, expect that interface to look different soon — the functionality isn't going away, it's just moving under one roof.

What's new in ChatGPT's voice mode?

Separately from Work and GPT-5.6, OpenAI also rebuilt voice mode entirely — and this update might affect more casual users than the Work agent does. Per OpenAI's official announcement of GPT-Live, the whole architecture changed: GPT-Live is built on a full-duplex architecture, meaning it can listen and speak at the same time.

If that sounds technical, here's the human version. The old Advanced Voice Mode worked like a walkie-talkie — it took turns speaking and listening in a rigid back-and-forth. Pause too long to think and it assumed you were finished. The new system fixes that specific annoyance: GPT-Live feels more like a real conversation because it can catch an interruption mid-sentence, hold space while you think, and signal it's listening with brief acknowledgments like "mhmm" or "got it" rather than going silent.

It's also smarter about knowing when to hand off. GPT-Live is OpenAI's smartest voice model yet — for questions that require web search, deeper reasoning, or more complex work, it delegates to the latest frontier model behind the scenes and brings the result back into the conversation when it's ready.

There's a plan split, though. GPT-Live-1 is locked to Go, Plus, and Pro subscribers, while free users get GPT-Live-1 mini, which has the same conversational architecture but lower capability. One current limitation to know about before you rely on it: at launch, GPT-Live does not support voice with video or screen sharing in ChatGPT, though OpenAI says it's working on adding those soon.

How do you know if you have it?

Look for a label change in the app. The biggest giveaway that you've got the upgrade is the Live label at the top of voice chats on mobile, and behind the ChatGPT drop-down on the web. If you'd rather use the old system for any reason, you're not locked out — you can still access legacy versions of ChatGPT Voice, including Standard and Advanced Voice Mode, where these features are available.

How should you actually start using this stuff?

Don't try to adopt everything at once — that's how people end up connecting a dozen plugins and never touching half of them. A more realistic rollout looks like this:

Check your plan tier first. If you're on Plus or above, go look for the Work tab before doing anything else. It won't show up for everyone immediately, and no amount of settings-digging will summon it early.
Start Work on a low-stakes recurring task. A weekly competitor scan, a Monday-morning dashboard summary, something you'd normally spend 20 minutes on. Don't hand it your entire quarterly report on day one.
Connect one plugin, not five. Slack or Google Drive is a good starting point since most people already have meeting notes and docs living there.
Try voice mode for actual back-and-forth thinking, not just dictation. The full-duplex behavior is most noticeable when you're talking through a messy idea out loud, not when you're issuing a single command.
Set a Scheduled Task for something you already do manually every week. That's the single highest-leverage use of ChatGPT Work right now — it's the difference between a tool you open and a system that runs itself.

What does this mean for how people actually work?

The pattern across all three launches — new models, a work agent, a rebuilt voice mode — points in one direction: OpenAI wants ChatGPT to stop being something you visit and start being something that's already working before you sit down. That's a genuinely different relationship with a piece of software, and it's worth taking seriously even if you only adopt one piece of it this month.

Pick the update that solves your actual bottleneck — not the one that got the loudest headline — and build a real habit around it before moving to the next one. The tools are moving fast enough already; you don't need to match their pace.