Openai updates the operator to O3, which makes its monthly subscription monthly monthly temptation

It was a big week for artificial intelligence advertisements after the events Microsoft, Google and Anthropic. But Openai ends matters with its own news. Not, we are not only talking About 6.5 billion dollars, its acquisition of the Jony IVE design team To lead a A new effort for devices, “IO” in Openai.

today , The company promoted its operator Independent web browser and an index control agent inside Chatgpt from the use of the previous GPT-4O language model to the latest and more powerful O3 Thinking Form.

The update, which has been released worldwide today, May 23, 2025, as a “research inspection” to pay subscribers in the Chatgpt Pro plan, which costs $ 200.

Basically, this is Openai’s way to say it is not a fully productive “sand” or perfect yet – may still have reservoirs and issues.

But with Google competitor offers AI’s higher subscription package at about $ 250 USD regularly (Currently it works to discount to $ 125 for the first three months) To reach the latest Gemini Multimodal, IMAGEN images, VEO Video Generation models, it seems that the Chatgpt Pro suddenly of Openaii suddenly is more expensive.

What is Openai and what is it?

The operator first appeared in January 2025 As an initial step for Openai in semi -independent factors, especially the computer using agents (CUAS). The idea is to bypass the Chatgpt Chatbot interface and allow the strong AI models from Openai to start taking more actions on the user.

Thus, the operator is designed to steer independently, click, pass and write to complete the tasks of web, such as reserving dinner reservations, collecting shopping lists, or ordering tickets for events. This allows the ability to complete the user’s tasks directly through the browser interface, from reservation to data collection online.

For safety, privacy and safety purposes, the operator did not use any web browser on the user’s computer or Mac. Instead, it was run in a default browser hosted by the cloud that can be accessed via an independent website-Operator.Chatgpt.com-where users can enter requests and monitor tasks to perform the agent in an actual time.

It has collected the capabilities of vision, thinking and interaction on the basis of GPT-4O, which represents a new direction for Openai in AIGNIC AI.

The product has been launched as a research inspection of Chatgpt Pro subscribers and integrated safety measures such as user assurances, monitoring mode, and restrictions on high -risk web platforms.

It was also tested in the contexts of institutions, including travel planning and civil services, which indicates its capabilities across both consumers and commercial environments.

O3 provides improved accuracy, structure and success rates

With this update, Openai aims to enhance performance across several main dimensions. The new O3 -based operator explains the improvement of stability and accuracy during browser reactions.

In practice, this means that it is likely to successfully complete the user’s tasks and with the least need to be corrected or repeated. Moreover, users can sign a clearer, more organized and more comprehensive responses.

In comparative assessments, the new model displays a distinctive preference for its predecessor. Human preference studies reveal that users prefer the O3 model for its style, comprehension and clarity. It also leads strongly to the following and efficient instructions, although the results of realistic right are more balanced between versions.

The performance of the third -party evaluation criteria reflects these improvements. on Osworld Standard This measures the completion of the browser -based tasks, and records the O3 42.9 model compared to 38.1 for the previous version.

However, Openai notes that due to the restrictions in the automated grades system, the actual performance gain can be closer to 20 percentage points!

On Webarena, the new model scored 62.9, an altitude of 48.1. The most dramatic improvement appears on GAIA criteria, where the O3 62.2 achieves, greatly exceeds 12.3 previous model.

Task comparisons are shown alongside these gains. In one of the examples that include a restaurant reservation request, the new model presented a clearer and more detailed list of reservations available, including sites, Michelin classifications, and sitting notes presented at a well -coordinated table. The previous version, despite functional, provided less information in a lesser -organized way, according to an included image with New O3 Player Notes notes:

The guarantees remain, as do general warning notes on use on sensitive financial transactions and access to the account

The O3 model also inherits safety measures that are presented with previous versions, with further control of its role as an agent system.

Openai has combined augmented training against the implementation of harmful tasks, weak injection, and errors that involve the user’s intention.

The assessments show that the model now confirms 94 % of sensitive procedures before implementing them, with 100 % confirmation of financial transactions. Immediate injection capacity has also decreased from 23 % to 20 %.

It is worth noting that the O3 operator maintains cautious limits on high -risk web interactions, such as e -mail or financial platforms, as the user’s supervision may require by placing the watch or refusing to follow up explicitly. These measures are part of the safety class approach that combines durability at the level of model and actual time monitoring.

While the promotion to the operator is a technical improvement, it also reflects Openai’s continuous commitment to spreading responsible artificial intelligence.

The system’s ability to take action in the real world provides new risks, and the development team continues to improve its safety protocols accordingly.

according to O3 -upThe model remains less than high -risk doorstep in categories such as biological and chemical misuse and has no local coding environment or crushing access, which reduces potential misuse vector.

The operator remains a research inspection and can only be accessed for Chatgpt Pro users. the API version of the operator You will continue to rely on the GPT-4O, at least at the present time.

The effects of the technical decision makers of the institutions

The promotion operator stands to enhance the functioning of professionals in artificial intelligence engineering, coordination, data management, and information technology security.

For those who build or maintain machine learning models, the accuracy of the improved model and structured outputs reduces the general expenses to verify the test validity and explore and repair errors.

In synchronization contexts, it provides a practical and reliable tool for automating the ingredients based on the browser for complex pipelines.

Data engineers can delegate manual web interactions-such as data verification and ignore-with more confidence, and free time to improve improvement at the higher level.

Meanwhile, security professionals are gaining a safer way to simulate the user’s behavior in checks and accidents response exercises, thanks to the safety mechanisms with typical layers.

Through these majors, the O player based on O3 provides ability to upgrade power and risk relief framework, which makes it a practical addition to the collection of modern technical tools.