On July 9 (local time), OpenAI unveiled ‘ChatGPT Work.’ It is an autonomous agent that takes on a task and handles it on its own for hours, returning finished materials like documents, sheets, and slides. A chatbot that used to only answer questions has now moved into a stage where it actually does the work for you. Coming right after last week’s GPT-5.6 model, this follow-up turns that model into a working tool — so today let’s walk through what ChatGPT Work is, how it works, how it’s priced, and how it changes enterprise adoption and the competitive landscape. 🤖

TL;DR

  • OpenAI announced ‘ChatGPT Work’ on July 9. Built on GPT-5.6, it pulls information from your connected apps and files, breaks work into steps and runs on its own for hours, and delivers the results as finished materials — sheets, slides, docs, and web apps.
  • Pricing is not a feature bundled into a flat subscription. Like the coding tool ‘Codex,’ it is charged by usage. Input, cached input, and output tokens each carry a different rate, so the longer it runs, the more it costs.
  • It opened first to the Pro, Enterprise, and Edu plans, and will expand to Plus and Business next. Enterprise safeguards — admin spend controls and limits on what agents can do — come with it.

🤖 What Makes ChatGPT Work Different — ‘Finished Output,’ Not ‘Conversation’

The heart of ChatGPT Work is that it delivers finished results, not answers. A traditional chatbot replied to questions with text, but ChatGPT Work takes a goal, pulls context from your apps and files, and produces reports and spreadsheets, presentations and documents — even shareable web apps. OpenAI described these outputs as “finished materials, not chat.”

It runs on GPT-5.6, formally released last week. OpenAI rolled the model out in three tiers — the flagship ‘Sol,’ the balanced ‘Terra,’ and the fast, low-cost ‘Luna’ — and ChatGPT Work is what connects that performance to actual work products. You can use it anywhere ChatGPT runs, including web, mobile, and desktop.

⏱️ It Works Alone for Hours — How the Autonomous Agent Operates

The biggest change is that a person no longer has to stay on top of it. ChatGPT Work breaks a complex project into small steps and carries it out on its own for hours. It pulls whatever information it needs along the way directly from your connected work apps and files. You provide only the ‘outcome’ you want, and the agent works through the path to get there on its own.

This approach resembles the flow of the coding assistant Codex. In fact, OpenAI explained that ChatGPT Work is “designed for longer, more involved work than a typical chat request” and follows the same usage structure as Codex. Several outlets characterized it as giving office workers the automation power of coding tools without the steep learning curve.

💳 The Pricing Feels Unfamiliar — Pay for What You Use, Not a Flat Fee

The part to watch closely is how it’s priced. ChatGPT Work is not simply a feature riding on your subscription; each run incurs a cost based on token usage. OpenAI said it bills input tokens, cached input tokens, and output tokens each at a different rate. The longer the agent works and the more material it handles, the higher the cost.

For reference, the API price of GPT-5.6, which powers ChatGPT Work, is (per 1M tokens) $5 input / $30 output for Sol, $2.5 / $15 for Terra, and $1 / $6 for Luna. Unlike the sense of using something without limit inside a flat fee, an autonomous agent is billed by the ‘unit of work’ — something worth weighing before adoption.

🏢 Enterprises Take Notice First — Admin Controls and Safeguards

It’s worth noting that control mechanisms built with enterprise adoption in mind arrived alongside it. OpenAI gives Enterprise and Edu admins spend controls. Workspace defaults, per-group limits, individual exceptions, and a request-and-approval flow for extra credits can all be managed in the admin console.

There are also safeguards that let admins limit what each agent can do within each app. Since an agent that moves on its own accesses real work systems, OpenAI appears to recognize that cost control and permission management are prerequisites for adoption. Access opened first on Pro, Enterprise, and Edu, and will expand to the Plus and Business plans in turn.

🥊 It’s Not Only OpenAI’s Story — The Opening of the ‘AI Agent Wars’

ChatGPT Work sits squarely in the industry-wide ‘agent competition.’ Around July 9, major AI companies released new models and tools one after another, shifting the axis of competition from simple conversational chatbots toward ‘AI that does the work for you.’ Elon Musk’s xAI unveiled ‘Grok 4.5’ the same day, and Meta is reported to have signaled an aggressive pricing strategy behind its latest model.

This trend shows the center of gravity in the AI industry moving from ‘model performance’ to ‘the work the model actually gets done.’ That said, how accurately and reliably autonomous agents finish tasks — and whether they can earn trust in real workplaces — still needs to be proven.

The Takeaway

The arrival of ChatGPT Work marks a turning point where AI moves from a ‘tool that answers’ to a ‘tool that works.’ Producing finished output by carrying on tasks by itself for hours, and shifting pricing from a flat fee to a usage basis to match, are the biggest departures from the previous generation of chatbots.

Three points to keep in mind. First, usage-based pricing carries a cost-management burden as much as a benefit; check your expected usage and set limits before adopting. Second, because autonomous agents access work systems, permission and security management become the core challenge of adoption. Third, with not just OpenAI but many companies moving in the same direction, the drive to test agent performance and reliability in real work is likely to continue through the second half of the year.

※ This post is for informational purposes only and is not investment advice.

Sources