Why is ChatGPT Agent Much Better Than Manus AI and GenSpark — Here is The Result
On July 18, 2025, OpenAI announced a new feature for ChatGPT called “ ChatGPT agent .” In a nutshell, it is a combination of two existing functions announced last year, “Operator” and “deep research,” which analyse and summarise information, and the AI operates a virtual computer to perform tasks based on the user’s instructions.
ChatGPT can now think and act, actively choose tools, and use its own virtual computer to complete tasks for you.
The era of Agent AI came a little earlier than we thought.
This release is the new ChatGPT Agent, which implements a key upgrade in the capabilities of the general agent.
Unlike previous large-scale basic model upgrades, the Universal Agent can automatically use a variety of tools for planning and help people complete complex tasks, including automatically navigating user calendars, generating editable PPTs, running code, and more.
The agent can connect to your Gmail, GitHub sites for information and troubleshooting, and use APIs to access a variety of applications. Agent’s AI intelligence has improved dramatically — the ChatGPT Agent-based model scored 41.6% on the HLE benchmark, almost double that of the o3 and o4-mini.
ChatGPT Agent is currently available to subscribers of the OpenAI Pro, Plus, and Team plans. Users who want to use ChatGPT can select “ChatGPT” in the drop-down menu of ChatGPT’s tools.
Sam Altman, CEO of OpenAI, said that watching the ChatGPT agent use a computer to perform complex tasks was a real “AGI” moment for me, seeing a computer think, plan, and execute makes all the difference.
Before we start! 🦸🏻♀️
If you like this topic and you want to support me:
like my article; that will really help me out.👏
Follow me on my YouTube channel
Subscribe to me to get the latest article.
ChatGPT can now use its own virtual computer to do the work for you, handling complex tasks from start to finish. Not only can users let ChatGPT perform requests such as “query annual financial reports,” but also intelligently browse websites, sift through results, prompt you to log in securely when needed, run code, conduct analysis, and even deliver editable slides and spreadsheets summarizing its research findings.
ChatGPT uses its own virtual computing environment to flexibly switch between inference and execution, handling complex workflows from start to finish based on the user’s instructions.
The most important thing is that the user is always in control. ChatGPT will ask for your permission before performing any important actions, and you can interrupt tasks, take over browsers, or stop working at any time.
For example, let the ChatGPT Agent search the City of San Francisco’s annual consolidated financial report
Another example is typing the prompt “I’m a tennis fan and want to go to Palm Springs to watch a tennis match, especially during the semi-finals/finals.
I live in San Francisco, please help me create a detailed three-day itinerary that includes flight schedules, hotel reservations, activities (races, hiking, food, spa, etc.). I loved hiking, vegan restaurants and spas. The total budget is $3000.
This itinerary needs to include: precise timing; the contents, fees and other details of each activity; If necessary, provide a link to purchase tickets or make a reservation. Then, let ChatGPT Agent help you create a detailed itinerary.
OpenAI said, “While ChatGPT Agent can already handle complex tasks, this release is just the beginning. We’ll continue to iterate and roll out major improvements regularly to make it more powerful, more useful, and more accessible to more users.”
Operator & evolution of deep research
In the past, Operator and deep research each had their own unique strengths: Operators were able to scroll, click, and type on web pages, while deep research was good at analysing and summarizing information.
However, the two play the greatest role in different scenarios, and each has its own areas of expertise. Operators can’t drill down or write detailed reports, and deep research can’t interact with web pages, further filter results, or access content that requires a user login.
OpenAI found that many of the tasks that users try to do with Operator are actually better suited to deep research, so they decided to combine the advantages of the two. By integrating these complementary capabilities into ChatGPT and introducing more tools, OpenAI unlocks entirely new capabilities in a single model.
It can now actively interact with the website, clicking, filtering, and collecting more precise and efficient results and seamlessly transition from natural communication to making specific action requests in the same conversation.
OpenAI has equipped ChatGPT Agent with a full suite of tools: This includes a visual browser that interacts with web pages through a graphical user interface, a text browser for handling simple inference web queries, a terminal (command line interface), and the ability to call APIs directly。
The agent can also use ChatGPT Connectors to connect apps like Gmail and GitHub, allowing ChatGPT to find information relevant to your prompts and use it in answers. Users can also log in to their account on any website by taking over the browser, helping it to go deeper and more extensively in terms of information retrieval and task execution.
All of this is done on the ChatGPT Agent’s own virtual computer. This preserves the contextual information needed for the task when using multiple tools.
ChatGPT Agent can choose to open a web page with a text browser or a visual browser as needed, download files from the Internet, run commands in the terminal to process the files, and then view the output results through the visual browser. At the same time, the strategy will be adjusted according to the task for fast, accurate and efficient execution.
ChatGPT Agent Vs Manus Vs GenSpark
In the field of AI agents, not only OpenAI but also startups and major companies around the world are competing fiercely.
Comparison with Manus
Manus is a self-driving type that leaves everything to the user, while ChatGPT Agent is a hybrid type where the human can take the brakes or the steering wheel as necessary. Manus minimises the burden on humans due to its high autonomy, but on the other hand, there is a problem that the internal process is difficult to see and difficult to verify.
Especially in important fields such as medicine, it has been pointed out that there is a risk that it is difficult for humans to later analyse “why it came to that” when Manus reach a conclusion, and that it is difficult to discover and correct even if an error occurs. ChatGPT Agent has an advantage in terms of accountability and transparency because the user can monitor it sequentially.
Manus was a hot topic when it was first released, with millions of people accessing it, and its sophistication has been proven, but it has been reported that the actual output quality is not always the best.
For example, in one comparative test, Manus achieved some results in a website auto-construction task, but was inferior in terms of the sophistication of the design and content, and was not as good as GenSpark
ChatGPT Agent does not currently have much data on such direct competition, but according to OpenAI’s own benchmarks, it is said to significantly exceed existing models in terms of completion rate and accuracy of complex tasks.
For example, it achieved 27.4% correct answers (significantly higher than conventional models) in the difficult mathematical formula problem set “FrontierMath, “ significantly surpassed human performance in the data science problem set “DSBench,” and recorded a 45.5% correct answer rate in spreadsheet manipulation tasks compared to Excel Copilot’s 20.0%
Comparison with GenSpark
GenSpark uses a “Mixture-of-Agents” architecture with 80+ specialised tools working as a team, plus unique visualisation of thought processes through graphical flowcharts. ChatGPT Agent uses a single model approach with text-only narration of progress, making GenSpark superior in explainability and process transparency.
GenSpark automatically generates slides, video clips, and multimedia content from a single instruction, rare among AI agents. ChatGPT Agent handles slides and images but lacks independent video creation, making GenSpark the winner for comprehensive multimedia output.
ChatGPT Agent achieved 68.9% in BrowseComp benchmarks (17.4 points higher than competitors) with flat-rate unlimited use through ChatGPT Plus.
GenSpark offers specialised UI/UX with $20/month + hourly charges, where one website generation consumes ~54 minutes (nearly half the monthly allowance), creating a cost barrier for heavy usage.
GenSpark beat Manus and Suna in comparative tests for stability and quality, especially for non-engineers building websites. However, ChatGPT Agent benefits from millions of existing users and OpenAI’s proven infrastructure, while GenSpark’s innovation comes with the risk of instability due to its smaller user base and newer technology.
What ChatGPT Agent Bring to the Table
The true value of ChatGPT Agent is that it can be used in combination with multiple functions, not just a single function. It creates a synergy effect where 1+1 becomes 3 or 4.
The first thing to note is the fusion of a web browser (visual) and text analysis (linguistic intelligence). ChatGPT Agent collects and understands information by switching between a visual browser and a text browser depending on the task. And the combination of code execution and web operations is also a major strength. For example, let’s say you ask the Agent to “extract prices from this product list (web page), perform statistical analysis, and graph the results.”
The Agent first opens the list page in a visual browser and scrapes the product names and prices. Up to this point, it is a browse operation similar to that of an Operator. Next, the data is passed to a virtual terminal, and statistical indicators of price distribution are calculated using Python and graphed using matplotlib
This series of multi-tool collaborations is carried out without human intervention. The Operator alone cannot execute programs, and conversely, Deep Research processes data using code, but the user has to set up API usage to collect data from the web.
Another thing we must not forget is the value of collaboration between users and agents. As agents become more advanced, it seems that human involvement is unnecessary, but ChatGPT Agent is designed to work with users. If the user says, “Let’s change course after all,” the agent will immediately correct the course and make use of the previous progress.
Conversely, the agent may ask, “I need more information to achieve my goal.” This two-way communication is a strength that fully autonomous AI does not have. If humans and AI complement each other in their areas of expertise, the quality of the output will be higher than if they were alone
summary
“ChatGPT Agent” is an innovative feature that has evolved ChatGPT from a conversational partner to a capable agent. Through its specific capabilities and features, and comparison with conventional technologies and competitors, we have seen the arrival of an era in which a single AI can seamlessly handle everything from information gathering to execution.
OpenAI has integrated the strengths of previous technologies, such as Operator and Deep Research, to bring this powerful Agent to the world with ease of use and safety in mind.
ChatGPT Agent has the potential to go beyond being a simple automation tool in terms of the fusion of conversation and action, and collaboration with humans. Specific usage scenarios showed many benefits, such as improving daily work efficiency and new service forms.
Reference :
🧙♂️ I am an AI Generative expert! If you want to collaborate on a project, drop an inquiry here or Book a 1-on-1 Consulting Call With Me.