Matthew Berman, a renowned AI enthusiast and commentator, recently unveiled an exciting new development from OpenAI that is set to transform how we interact with the internet and AI: the ChatGPT Agent. By combining the strengths of OpenAI’s Operator, Deep Research, and ChatGPT, this innovative agentic system offers a powerful, streamlined way to perform complex web-based tasks autonomously. In this article, we’ll dive deep into what ChatGPT Agent is, how it works, its impressive benchmarks, and the implications of this groundbreaking AI technology.
Table of Contents
- 🚀 What is ChatGPT Agent?
- 🧩 How ChatGPT Agent Works: A Closer Look
- 📊 Benchmarking ChatGPT Agent’s Performance
- ☁️ Vultr: Powering the Future of AI with Cloud Infrastructure
- ⚠️ Risks and Considerations with ChatGPT Agent
- 🌐 The Changing Role of Humans on the Internet
- 🔍 Detailed Demonstrations: What ChatGPT Agent Can Do
- 📈 What’s Next for ChatGPT Agent and AI Agents in General?
- ❓ Frequently Asked Questions (FAQ)
- 🔚 Conclusion
🚀 What is ChatGPT Agent?
OpenAI’s ChatGPT Agent is a unified system that integrates three major breakthroughs in AI tooling:
- Operator: The ability to interact directly with websites, including clicking, navigating, and performing tasks just like a human user would.
- Deep Research: The skill to synthesize information from multiple sources over long periods, allowing for in-depth research and comprehensive reporting.
- ChatGPT: The conversational fluency and intelligence that powers natural, human-like interactions.
By merging these capabilities, OpenAI has created an agent that can autonomously browse the web, extract and organize information, perform complex multi-step workflows, and even create documents such as reports or spreadsheets. It is like having a virtual assistant that can not only chat but also take real-world actions on your behalf.
Matthew demonstrated the agent’s power by showing it booking a dog-friendly Hipcamp with a private hot tub near San Francisco. Unlike previous iterations, where you watched the operator click through websites with visible trial and error, ChatGPT Agent offers a sleek, simplified user interface. It searches, reads, and reasons through websites behind the scenes, presenting users with clean, concise results after a chain of thought lasting several minutes.
🧩 How ChatGPT Agent Works: A Closer Look
At its core, ChatGPT Agent runs on a virtual computer environment similar to Manus, another AI agent platform. This virtual machine allows the agent to execute commands, navigate multiple websites, and manage data simultaneously. Manus even acknowledged ChatGPT Agent as a notable competitor in the agent space, sparking interesting comparisons between the two.
When you activate ChatGPT Agent (available in ChatGPT Pro Plus and Team accounts), you can select from various example queries such as:
- Researching carbon capture cost trends and projections
- Retrieving WHO health professional density data by profession
- Booking accommodations with specific features (e.g., dog-friendly with hot tubs)
- Organizing vegetarian recipes based on protein efficiency
Unlike previous operator tools that visually showed you every click and mistake, ChatGPT Agent streamlines the process with a scrubber timeline interface. You can fast-forward or rewind to see each step the agent took over a 12-minute reasoning period, including searches, website navigation, and data extraction.
For example, when booking the Hipcamp stay, ChatGPT Agent:
- Loaded the Hipcamp website
- Clicked through calendar dates to check availability
- Selected the number of guests
- Verified all details before presenting the best options
This autonomous interaction mimics a human’s browsing and decision-making but with superhuman patience and speed.
📊 Benchmarking ChatGPT Agent’s Performance
One of the most exciting aspects of ChatGPT Agent is its performance on various AI benchmarks, especially those designed to test multi-agent and tool-using AI systems.
Matthew shared results from “Humanity’s Last Exam,” a benchmark measuring AI’s ability to complete economically important, real-world tasks compared to humans. The results showed:
- ChatGPT Agent (with browser + computer + terminal tools): 41.6% success rate
- Deep Research alone: 26%
- OpenAI GPT-3 with Python + browsing: 24.9%
- ChatGPT Agent with no tools: Lower than with tools, underscoring the importance of tool integration
For context, the recently released Grok 4 Heavy scored 44.4%, slightly higher than ChatGPT Agent. However, Grok 4 is a brand-new model, while ChatGPT Agent is built on existing models customized for agentic tasks.
Other benchmarks included Frontier Math, where ChatGPT Agent achieved 27.4%, outperforming older models like GPT-4 Mini and GPT-3. On DS Bench, which tests data science skills such as data analysis and modeling, ChatGPT Agent outperformed humans with scores of 89.9% in some tasks compared to humans’ 64.1%.
Spreadsheet Bench, however, revealed areas for improvement. While humans scored 71.3% at creating and editing spreadsheets, ChatGPT Agent with Excel access scored 45.5%, indicating that there is still some way to go for AI to fully match human proficiency in spreadsheet tasks.
☁️ Vultr: Powering the Future of AI with Cloud Infrastructure
Matthew’s video was sponsored by Vultr, the world’s largest independent cloud provider, which offers robust GPU provisioning that is ideal for AI projects. Vultr’s infrastructure spans 32 locations across six continents and offers the latest AMD and Nvidia GPUs, ensuring low latency and top-tier performance for AI developers.
Some highlights of Vultr’s offerings include:
- Industry-leading price-to-performance ratio
- Accessibility and reliability for both hobbyists and enterprises
- Global, composable cloud infrastructure that helps avoid vendor lock-in
- Kubernetes engine support for scaling beyond single containers
For those eager to experiment with AI or scale production workloads, Vultr provides a compelling platform. Matthew’s viewers can get $300 in credits for the first 30 days by visiting getvultr.com/forwardfutureai and using promo code BERMAN300.
⚠️ Risks and Considerations with ChatGPT Agent
While ChatGPT Agent’s capabilities are impressive, they come with new risks. Because the agent can take direct actions on the web and work with your personal data, security and privacy concerns are paramount. Malicious actors could potentially exploit the system to extract sensitive information if users are not cautious.
Matthew cautions users to be very careful about what information they share with ChatGPT Agent. The AI’s ability to interact autonomously means that if prompted cleverly, it could inadvertently disclose personal or sensitive data. This introduces a new layer of risk that users must manage carefully.
🌐 The Changing Role of Humans on the Internet
One of the broader implications Matthew highlights is the evolving relationship between humans and the internet. As AI agents like ChatGPT Agent take on more of the heavy lifting—searching, filtering, synthesizing, and acting—humans may become more detached from the raw internet experience we know today.
On one hand, this is exciting because it frees users from sifting through vast amounts of low-quality or irrelevant information. Assigning an agent to handle time-intensive research or booking tasks saves effort and improves efficiency.
On the other hand, entrusting AI agents as gatekeepers raises questions about trust, transparency, and control. How do we ensure the agent filters information fairly and without bias? How do we maintain autonomy over decisions when an AI acts as an intermediary? These are open questions as the technology continues to mature.
🔍 Detailed Demonstrations: What ChatGPT Agent Can Do
Matthew’s demonstration videos highlight several concrete use cases that show ChatGPT Agent’s versatility:
Booking Dog-Friendly Hipcamp with a Hot Tub Near San Francisco
The agent autonomously navigated the Hipcamp website, searching for listings with a private hot tub suitable for dogs. It clicked through calendar dates, verified availability for the selected dates, and filtered results based on the number of guests. After about 12 minutes of reasoning and web interactions, it presented a detailed summary of the best options, including Kings Mountain Camps’ apartment farm stay in Woodside.
Organizing Vegetarian Recipes by Protein Efficiency
In another example, ChatGPT Agent was tasked with organizing vegetarian recipes according to their protein content and efficiency. It performed multiple web searches, read through recipe data, and compiled the information into a structured spreadsheet. This showcases how the agent can perform data-intensive synthesis tasks and produce actionable outputs.
📈 What’s Next for ChatGPT Agent and AI Agents in General?
ChatGPT Agent represents a significant step toward AI systems that can act autonomously on the web, combining browsing, computation, and conversation. As these agents improve, especially when paired with next-generation models like GPT-5, their intelligence and usefulness could skyrocket.
We can anticipate AI agents becoming indispensable tools for professionals, researchers, and everyday users alike, handling complex workflows, managing data, and interacting with digital ecosystems in ways that were previously impossible.
However, the need for robust safety measures, transparency, and user control will only grow. The balance between convenience and privacy, autonomy and oversight, will define the future of AI agents.
❓ Frequently Asked Questions (FAQ)
What is ChatGPT Agent?
ChatGPT Agent is a new AI system from OpenAI that combines the interactive capabilities of Operator, the research prowess of Deep Research, and the conversational intelligence of ChatGPT to autonomously perform complex web-based tasks.
How does ChatGPT Agent differ from previous AI tools?
Unlike prior tools that showed visible browsing and trial-and-error, ChatGPT Agent operates with a streamlined interface, performing searches, navigation, and reasoning behind the scenes. It can also create outputs like reports and spreadsheets autonomously.
Who can access ChatGPT Agent?
ChatGPT Agent is available to users with Pro Plus and Team accounts on ChatGPT, meaning you don’t need the most expensive tier to use it.
What kinds of tasks can ChatGPT Agent perform?
The agent can handle multi-step tasks such as booking accommodations, conducting detailed research, retrieving data, and organizing information into spreadsheets or reports.
How well does ChatGPT Agent perform compared to humans?
Benchmark tests show ChatGPT Agent outperforming humans in some data science tasks and economically important tasks about 30% of the time. However, in spreadsheet editing, humans still outperform the agent.
What are the risks of using ChatGPT Agent?
Because the agent can interact with websites and handle personal data, there is a risk of data leakage or exploitation by malicious actors. Users should be cautious about what information they provide and how they use the agent.
How does Vultr relate to ChatGPT Agent?
Vultr is a cloud provider sponsoring Matthew’s video. They offer GPU provisioning ideal for AI projects, which can be used by developers building or scaling AI agents like ChatGPT Agent.
What does the future hold for AI agents?
AI agents will likely become increasingly intelligent and autonomous, capable of managing complex workflows and interfacing with the internet on behalf of users. This evolution will require careful attention to safety, trust, and user control.
🔚 Conclusion
ChatGPT Agent is a fascinating glimpse into the future of AI—where agents don’t just chat but act autonomously, navigating the web, synthesizing data, and performing meaningful tasks. Matthew Berman’s exploration highlights both the potential and the challenges of this technology. As AI agents become smarter and more capable, they promise to revolutionize how we work, research, and interact with digital content.
Yet, this power comes with responsibility. Users must remain vigilant about security and privacy risks, and developers need to prioritize transparency and control. The internet as we know it may soon transform into an AI-mediated experience, and ChatGPT Agent is leading the charge.
For those eager to experiment with AI or scale up their projects, cloud providers like Vultr offer the infrastructure needed to power the next generation of AI applications. With $300 in credits available via Matthew’s promo code, it’s a great time to dive in.
ChatGPT Agent’s wild capabilities are just the beginning. Stay tuned as AI continues to reshape our digital landscape in ways both exciting and profound.
