OpenAI is upgrading the AI model that powers Operator, its autonomous agent capable of browsing the web and using software within a cloud-hosted virtual machine to fulfill user requests.
The new version will utilize a model based on o3, one of OpenAI’s latest “reasoning” models, replacing the previous custom version based on GPT-4o. The o3 model is significantly more advanced, especially in tasks involving math and reasoning.
In a blog post, OpenAI stated, “We are replacing the existing GPT-4o-based model for Operator with a version based on OpenAI o3.” The API version of Operator will continue to run on GPT-4o.
ICYMT: The Ultimate Boys’ Trip: 7 Must-Visit Destinations Before 40
Operator is one of many advanced agentic tools introduced by AI companies recently, as the race intensifies to create sophisticated agents capable of performing tasks with minimal supervision. For instance, Google offers a “computer use” agent through its Gemini API, which can also browse the web and take actions for users, along with a consumer-focused tool called Mariner. Anthropic’s models similarly perform computer tasks, such as opening files and navigating web pages.
According to OpenAI, the new model, known as o3 Operator, has been “fine-tuned with additional safety data for computer use,” including datasets designed to clarify the model’s decision-making boundaries regarding confirmations and refusals.
OpenAI has released a technical report detailing o3 Operator’s performance in safety evaluations. Compared to the GPT-4o model, o3 Operator is less likely to refuse requests for “illicit” activities and is also more resilient against prompt injection attacks.
OpenAI noted, “o3 Operator uses the same multi-layered approach to safety that we used for the 4o version.” While it inherits advanced coding capabilities, o3 Operator does not have native access to a coding environment or terminal.
SOURCE: TECH CRUNCH
