ChatGPT Work: Your AI colleague

OpenAI has spent three years teaching ChatGPT to answer questions. Today, it announced the model is now meant to finish your workload instead.

The company launched ChatGPT Work, a new mode inside ChatGPT built to take on multi-step projects across a person’s apps and files rather than responding to a single prompt and stopping there. It arrives alongside GPT-5.6, OpenAI’s newest model family, which now powers both Work and Codex, the company’s coding agent.

What ChatGPT Work actually does:

Gathers information across a user’s connected apps and workflows to build finished materials: documents, spreadsheets, slide decks, reports, and interactive web apps
Stays on complex projects for hours by breaking them into smaller steps and working through them independently
Lets a user state an outcome, attach the relevant files or apps, and leave the model to find its own path there
Runs on GPT-5.6, which OpenAI says is now state of the art at reasoning through multi-step tasks and matching a user’s existing templates and reference files

In OpenAI’s own words, ChatGPT is becoming “a partner for your most ambitious work” rather than a place to ask quick questions. The company frames Work as the point where Codex’s agentic technology, originally built purely for coding, finally reaches everyday knowledge workers. More than five million people now use Codex every week, and more than one million of them already use it for tasks that have nothing to do with software development, a signal OpenAI reads as proof that people want an agent that does the work, not just one that writes code.

Built for the whole day, not just the desk

ChatGPT Work at work — ChatGPT Work won’t be static but follow you around.

Photo: OpenAI

Work is designed to follow a person around rather than remain pinned to a single device. A task can start on a phone during a commute, get checked mid-afternoon between meetings, and get picked back up on the web once someone’s back at their desk. When on the desktop, it can access local files and installed apps directly, and comes with a built-in browser so it can pull in live websites, tools, and online files without a user tabbing out.

Two features do most of the heavy lifting:

Plugins connect ChatGPT to the tools where work already happens. These include Slack, Microsoft Teams, Gmail, Google Drive, SharePoint, calendars, CRMs and project trackers. Typing “@” followed by an app’s name in a prompt tells ChatGPT exactly where to pull context from, and a new unified plugins directory brings all connections into one place.
Scheduled Tasks let a person hand off recurring administrative tasks entirely. For example, reviewing new Slack messages each week to refresh a meeting agenda, checking a dashboard every morning and sending a summary, or updating a presentation the moment new feedback lands by email. Nothing runs unsupervised without a user’s say-so. People choose what ChatGPT can access, when it should check in, and which actions need explicit approval before going ahead.

OpenAI is also introducing Sites in public beta, a way to turn a piece of work straight into an interactive dashboard, tracker, or internal tool, shareable by URL and testable inside ChatGPT itself before it goes live.

The early evidence

Using a browser to access ChatGPT Work — When using ChatGPT Work via a browser

Photo: OpenAI

OpenAI shared four customer accounts from its testing period, and the specifics are worth sitting with rather than skimming. At Zapier, Angela Ferrante, Head of Enterprise Marketing, used Work to build a repeatable system for reviewing thousands of leads a month, tracing customer touchpoints across the CRM and email, flagging where follow-ups broke down, and generating a weekly executive dashboard that ended up surfacing seven figures in missed pipeline.

At RingCentral, Vaneet Seth, R&D Efficiency Manager, turned a manual monthly launch check into an automated workflow that reviews release plans, Jira tasks and go-to-market schedules, then produces source-backed reports naming exactly who owns what. The result: Seth went from supporting a single product manager to supporting roughly 50.

Virgin Atlantic’s Nathan Bolt, Head of Digital Products, fed Work a customer journey and a list of competing airlines while building the airline’s five-year plan, then had it research and assess where Virgin led or lagged. What would normally be weeks of manual competitive analysis came back as a dataset his team could review and refine in hours.

And at NVIDIA, Will Daney, Go-to-Market Manager, used Work to automate prep for GTC, the company’s global conference, a task that previously ate up around 40 percent of his pre-event time in spreadsheets. ChatGPT tracked account registrations, meeting plans and field sales prep, then synthesised hundreds of session transcripts and customer-meeting notes afterwards so the team’s two-week review became a discussion of findings rather than a data-assembly exercise.

Within OpenAI, the company says nearly every team, including finance and sales, now uses Work and Codex daily. One sales example occurred when a discovery call turned into a tailored proof of concept for a mission-critical customer problem within 24 hours, a process that normally takes weeks, freeing the salesperson to stay focused on the customer relationship rather than the paperwork behind it.

The model underneath: GPT-5.6

Creating tasks to run automatically. — You can schedule tasks to be done easily.

Photo: OpenAI

GPT-5.6 launches in three tiers, Sol (the flagship), Terra (a balanced everyday model), and Luna (the fastest and cheapest of the three). OpenAI is leaning hard on a performance-per-dollar argument here rather than raw capability alone:

On Agents’ Last Exam, a benchmark covering long-running professional workflows across 55 fields, Sol scores 53.6, ahead of rival model Claude Fable 5 by 13.1 points
On the Artificial Analysis Coding Agent Index, Sol sets a new state of the art at 80, while using under half the output tokens and time, and roughly a third less estimated cost than Fable 5
Terra and Luna, the cheaper tiers, reportedly outperform Fable 5 on the same benchmark at around one-sixteenth the cost
A new “ultra“ setting coordinates four AI agents working in parallel by default (scalable to 16) to push through especially demanding tasks faster

Early testers quoted in OpenAI’s own materials back the efficiency claim from several angles. Notion’s Simon Last called Sol “the most tenacious problem-solver we’ve seen yet,” while Cognition’s Scott Wu described it as combining “strong coding-agent performance with very strong cost efficiency.” Canva’s Danny Wu said it was around 1.6 times more token-efficient on slide creation specifically, and Microsoft’s Charles Lamanna confirmed GPT-5.6 is now the preferred model inside Microsoft 365 Copilot.

The model also showed real gains in cybersecurity and life sciences research, alongside what OpenAI calls its most robust safety system yet, including a dedicated Trusted Access program for verified cybersecurity professionals and a reasoning-based monitor layered on top of standard classifiers to detect harmful use in context.

Getting started, and what happens to the old app

Codex can be consider the predecessor to Work. — Those currently using Codex will be upgraded to Work automatically.

Photo: OpenAI

Work begins rolling out today on web and mobile for Pro, Enterprise and Edu plans, with Plus and Business following within days. On desktop, the story is simpler: the updated ChatGPT app, combining Chat, Work and Codex in one place, is available globally on Mac and Windows starting now, and free of charge to Free-tier users too, a notably more generous availability window than the paid-only rollout on web and mobile.

Anyone currently using the standalone Codex desktop app will be upgraded automatically. The current ChatGPT desktop app is being renamed to ChatGPT Classic and will remain installed and usable, but won’t receive further updates.

According to OpenAI, this is a first step rather than a finished product, describing its ambition as pushing ChatGPT “beyond answering questions to helping everyone turn their biggest ideas into reality.” Given how quickly the assistant-to-agent shift has moved across the industry this year, from Google’s Gemini Spark to Microsoft’s own Copilot agents, ChatGPT Work seems less novelty feature and more a step forward to where the whole category is heading next.

Read Full Article At Source