ChatGPT Work launches as OpenAI targets workplace tasks

OpenAI has launched ChatGPT Work, an agent inside ChatGPT designed to carry out multi-step jobs across apps, files and the web. The pitch is workplace software: a system that can keep moving through documents, browser tabs and connected services after a worker sets the task.

OpenAI said Pro, Enterprise and Edu users get first access on web and mobile, while Business and Plus accounts will be added over the next few days. That order matters. It puts the product in front of paying teams and gives IT departments an early look at whether a ChatGPT agent can fit into daily work without creating fresh permission, logging and audit headaches.

OpenAI is backing the workplace push with Codex usage data. More than 5 million people use Codex each week, the company said, and more than 1 million use it for work outside software development. The claim is straightforward: if workers already trust an AI system with code and adjacent tasks, they may also hand it research, document handling and other office jobs. Safety and tool access will decide how far that goes.

From chat to workflow software

Bloomberg reported ChatGPT Work is meant to handle a broader range of complex tasks for hours at a time. That is the business-professional angle OpenAI is chasing. The customer is a worker who wants software to stay on a job after the first prompt. OpenAI paired the launch with GPT-5.6, tying the agent story to its latest model release rather than treating the two as separate announcements.

Reuters reported the release intensifies the workplace AI race, with Anthropic, Microsoft and Google all chasing corporate budgets. The contest is less about who has the neatest chatbot window and more about which tool can act across files, browser sessions and internal systems without breaking company rules.

For Australian organisations, the practical questions are already on the table. Teams trialling Microsoft Copilot, Gemini and Claude-based tools will judge ChatGPT Work on permissions, data access, audit trails and workflow fit. Governance staff will want controls for file access and browser activity before the agent reaches broader staff use.

Paid tiers first

The first-wave rollout shows OpenAI starting where the money and risk sit. Pro and Enterprise users already pay for higher-end features. Edu access brings in universities and research institutes, including in Australia, where shared systems, research data and compliance rules can slow approval for autonomous tools.

Business and Plus accounts come next. That gives OpenAI a small feedback window with heavier users before a wider subscriber rollout, and it lets the company sell ChatGPT Work as an upgrade inside existing plans rather than a separate software line.

On the model side, OpenAI said on its GPT-5.6 page the companion model scored 53.6 on Agents’ Last Exam, an agent-focused benchmark. Benchmarks do not settle enterprise buying decisions. They do help OpenAI argue that longer-running work products now rest on models built for planning, tool use and follow-through.

The harder proof will come inside customer workflows. ChatGPT Work has to handle company data reliably, show what it did, and stop when permissions say stop. For local CIOs and university IT teams, those controls will matter more than launch-day claims when they decide which assistant gets access to staff workflows first.