OpenAI has unveiled a novel reasoning model called o3-pro, which the organization asserts is its most advanced to date.

On Tuesday, the developer of ChatGPT announced o3-pro on X, outlining its improvements over o3. OpenAI highlighted enhanced performance in “science, education, programming, data analysis, and writing” and mentioned that reviewers scored it higher for “clarity, comprehensiveness, instruction-following, and accuracy.”

This Tweet is currently unavailable. It might be loading or has been removed.

o3-pro is available via the API and to ChatGPT Pro and Team users, with Enterprise and Edu access set to commence next week.

OpenAI’s o3-pro represents the latest iteration in its line of reasoning models, which break down tasks into sequential steps for potentially improved and dependable responses, contrasting with conventional LLMs. In this regard, reasoning models like o3-pro are considered superior for intricate tasks, and OpenAI claims that o3-pro “excels at math, science, and coding.” OpenAI published benchmark assessments indicating that o3-pro surpasses o1-pro and o3 in these domains.

Researchers at Apple recently discovered notable limitations in reasoning models, including OpenAI’s o3-mini. When faced with the traditional Tower of Hanoi challenge, which entails relocating discs among pegs, the models stumbled as complexity intensified and ultimately halted attempts despite having sufficient computational resources.

This research sparked conversations within the AI community as it proposed that reasoning models might not be as sophisticated as asserted. Nevertheless, experts pointed out that the study concentrated on a specific issue and did not evaluate human and reasoning model efficacy on the same tasks. Therefore, it is not definitive that AI tools are wholly exaggerated, but the results suggest they may not be optimally suited for every type of task.

o3-pro can utilize tools within ChatGPT, such as web search, Python, visual document analysis, and tailored responses with memory capabilities. However, it does not facilitate image generation or canvas, OpenAI’s project interface, according to the release notes. Responses are also expected to take more time since o3-pro can access tools.

Disclosure: Ziff Davis, parent company of Mashable, filed a lawsuit against OpenAI in April, alleging copyright infringement in the training and functioning of its AI systems.