OpenAI is updating its AI assistant, Operator, based on a new, more powerful o3 model. This new version is much smarter than the previously used GPT-4o model, and is especially superior in logical solutions and calculations.
According to an OpenAI blog post, the Operator can now browse the Internet to carry out user tasks, and can independently launch some programs on a remote computer. The updated model has been trained based on special data on safe computer use.
What’s changed?
- The Operator is now based on the o3 model.
- Unlike GPT-4o, this model has been found to execute fewer dangerous or malicious requests.
- It is also more resistant to AI attacks such as prompt injection.
- The o3 Operator still knows how to write code, but it does not connect directly to a terminal.
AI companies are currently trying to outdo each other. For example, Google has released a “computer assistant” through its Gemini API, while Anthropic has released agents that can open files and navigate web pages. OpenAI, on the other hand, considers security a priority: “o3 Operator provides security through a multi-layered approach,” the company says in a statement.
