OpenAI releases o3-pro, an enhanced version of its o3 AI reasoning model

OpenAI releases o3-pro, outperforming previous AI models in multiple dimensions.

: OpenAI has introduced o3-pro, an enhancement of its earlier o3 reasoning AI model, made available for ChatGPT Pro, Team users, and the developer API. The model operates at a cost of $20 per million input tokens and $80 per million output tokens, significantly improving clarity and accuracy over its predecessor. Despite its strengths, it faces limitations such as slower response times and lack of image generation capability. Nevertheless, o3-pro surpasses Google Gemini 2.5 Pro in AIME 2024 and Anthropic's Claude 4 Opus in PhD-level science assessments.

OpenAI has recently launched o3-pro, a refined iteration of its o3 AI reasoning model. Designed to approach problem-solving in a systematic and stepwise manner, o3-pro is intended to perform effectively in domains such as physics, mathematics, and coding. This advanced model is considered a significant upgrade to its predecessors, capable of offering more controlled and reliable responses, which are particularly critical in specialized fields involving intricate subject matter.

Existing users of OpenAI such as those with ChatGPT Pro and Team accounts, now have access to o3-pro. It replaces the older o1-pro model and promises to bring enhancements in performance aspects like instruction-following, clarity, and accuracy. Its deployment began for these user groups on a Tuesday with Enterprise and Edu users anticipated to receive access a week later. OpenAI has also integrated the new model into its developer API.

Financially, utilizing o3-pro comes with a significant cost structure. OpenAI charges $20 for every million input tokens and $80 for each million output tokens processed via the API. This pricing scheme reflects the advanced capabilities embedded within o3-pro, which are set to exceed expectations particularly in mathematical problem-solving, as demonstrated by its superior performance in benchmarks like AIME 2024 and GPQA Diamond.

However, the model is not without its challenges. OpenAI disclosed certain functional restrictions such as longer response times when compared to the older o1-pro model. Additionally, the capability to generate images is absent, and there is ongoing work to address technical issues related to chat functionalities. Furthermore, the AI workplace application Canvas remains unsupported by o3-pro.

Notwithstanding the hurdles, o3-pro has been positively received due to its analytical prowess and versatility, especially in science and education sectors. It employs advanced tools which enable functionalities like Python scripting, web searching, and visual input analysis. On major AI benchmarks, o3-pro has outstripped renowned AI models, including Google's Gemini 2.5 Pro in specialized math tests and Anthropic's Claude 4 Opus in technical science evaluations. Such achievements affirm OpenAI's momentum in advancing AI research and application.

Sources: TechCrunch, OpenAI

AIGenerative AIscience