OpenAI ChatGPT Work launch: GPT-5.6, Scheduled Tasks and pricing

OpenAI released ChatGPT Work today, a new ChatGPT mode built to handle longer, multi-step projects and run for hours if needed. The company says the tool can “stay with a project for hours if needed, and turn a goal into finished work,” and it adds Scheduled Tasks, deeper desktop access, and broad plugin integrations to try to automate end-to-end workflows.

What is ChatGPT Work and how does it work?

ChatGPT Work is a ChatGPT view designed for extended, involved work rather than short chats, and it can access outside apps automatically when needed. It connects to workplace tools such as Slack, Microsoft Teams, Google Drive, and SharePoint via custom plugins, and users can force a specific third-party tool with an “@” reference. The desktop client can access and modify local files and includes a built-in browser for online resources. An updated ChatGPT Chrome extension will let users perform web-native tasks without switching apps, and OpenAI says its Atlas web browser is being sunset.

The release follows earlier experiments with an “Agent Mode” in OpenAI’s Atlas browser, which testers found would often stop after a few minutes, limiting usefulness for complex work. ChatGPT Work claims to resolve that by running for much longer and by adding Scheduled Tasks, a more capable take on cron jobs that can run on schedules or when monitored events occur and be monitored from a phone.

How will ChatGPT Work be billed and what models power it?

OpenAI will bill ChatGPT Work using the same structure as its Codex product, with subscription plans that range up to $100 a month and built-in usage limits governed by a credit system. The company warns that complex, longer-running tasks “may use more of your plan’s included usage.” For larger customers, ChatGPT Enterprise and Edu subscribers can set overall spend limits or group- and individual-level limits to control high-impact work.

The rollout coincides with a new GPT-5.6 model that OpenAI says delivers “stronger performance per dollar” for demanding tasks. GPT-5.6 can operate in three tiers; the most expensive, high-performance tier costs $5 for a million input tokens and $30 for a million output tokens.

OpenAI also folded its coding-focused Codex app into ChatGPT Work. Codex technology is now built into ChatGPT Work while remaining available as a separate “view” inside the ChatGPT app. The existing ChatGPT desktop app is being rebranded as “ChatGPT Classic,” with the basic conversational experience relegated to a quick-chat option.

Why it matters

ChatGPT Work pushes the product from short, conversational queries toward sustained automation. That changes the risk and reward calculus: a tool that can run for hours and manipulate files, apps, and scheduled jobs can remove tedious human steps, but it also increases exposure to data access, accidental changes, and cost overruns if left unchecked. OpenAI provides enterprise controls and a Compliance API to limit app and file access, and spend caps for Enterprise and Edu customers, but the product design still centralizes broad access to workplace systems.

The pricing signals matter too. A high-performance GPT-5.6 tier priced at $5 per million input tokens and $30 per million output tokens makes prolonged agent-driven work potentially expensive, especially since OpenAI explicitly warns that extended tasks may consume substantial plan usage. That pricing structure will shape which teams use persistent agents and how they budget for them.

What to watch

Watch whether scheduled automation actually runs reliably for hours and whether enterprise controls prevent unwanted access or runaway spend. Also watch adoption signals: who uses the top GPT-5.6 tier and how organizations apply group- and individual-level spend limits to contain cost.