ChatGPT agent now thinks and acts with its own virtual computer

OpenAI has officially launched its most powerful AI assistant yet: ChatGPT agent. Unlike previous versions that simply responded to queries, this new agent is capable of completing real-world tasks from start to finish—using its own virtual computer.

Introducing a new era of intelligent AI assistants

This means ChatGPT can now think, act, and deliver—from creating editable spreadsheets and PowerPoint slides to making restaurant reservations and conducting deep competitive research.

What makes ChatGPT agent different?

At the heart of this new feature is a blend of OpenAI’s most advanced tools: the Operator, which can browse and interact with websites; and Deep Research, which synthesizes large volumes of data. Now fused into one unified system, the agent can click, scroll, run code, summarize findings, and even create documents—all without switching contexts.

For example, you could ask it to "analyze three competitors and create a slide deck," and the agent will deliver editable PowerPoint slides that summarize its insights.

Real-world applications already in action

ChatGPT agent is not just a gimmick—it’s already proving useful in real business scenarios. Product manager Neel Ajjarapu shared that it's capable of handling *first- and second-year level financial analysis,* tasks that might have once taken hours but now take minutes.

It’s also able to assist with:

  • Booking restaurants or events
  • Creating amortization schedules
  • Writing competitive industry reports
  • Planning personal events like weddings or travel itineraries

Collaborative and adaptive workflows

One of the most powerful aspects of ChatGPT agent is its adaptability. Users remain in control at all times. You can pause tasks, adjust instructions mid-process, or even take over the browser if needed.

“I would explain this to my own family as cutting edge and experimental; a chance to try the future,” said OpenAI CEO Sam Altman, urging caution in high-stakes or sensitive tasks.

Safety measures and privacy-first design

With such power comes responsibility. OpenAI has built in several layers of safeguards: - *Explicit user confirmation* for tasks with real-world consequences - Watch mode for tasks like sending emails - Refusals for high-risk actions such as bank transfers.

Privacy is also a top priority. For example, when using browser takeover mode, the model never stores passwords or personal inputs, and you can clear all browsing data with one click.

Benchmarks show superior performance

OpenAI rigorously tested ChatGPT agent across a wide range of benchmarks:

  • Humanity’s Last Exam: new SOTA score at 41.6
  • SpreadsheetBench: 45.5% accuracy vs. Excel Copilot’s 20.0%
  • DSBench (Data Science): outperformed humans significantly
  • FrontierMath: highest known scores in complex problem-solving
  • BrowseComp: 68.9% score, well ahead of previous models

Available now—if you’re subscribed

ChatGPT agent is rolling out to Pro, Plus, and Team users starting today. Enterprise and Education users will follow later this summer. Pro users will get up to 400 messages per month, with flexible credit-based add-ons.

Unfortunately, it's not yet available in the European Economic Area or Switzerland.

Looking ahead: more polish and power

While the slideshow and spreadsheet functions are already game-changers, OpenAI admits they’re still in beta. The company is actively improving the formatting, usability, and ability to import and edit existing files.

But even in its early stages, ChatGPT agent is a major leap forward. It redefines what it means to have a digital assistant—not just answering questions, but doing the work for you.

Comments

Popular posts from this blog

5 Portable USB Gadgets That Make Everyday Life More Convenient

Nothing’s Design-First Vision: How Carl Pei Is Reimagining Consumer Technology

TSMC pushes chip innovation forward without relying on costly new ASML machines