OpenAI has launched its latest AI reasoning model, named o3-pro, providing an upgrade from the previous o1-pro version. This advanced model is now available to users of ChatGPT Pro and Team services, with the rollout to Enterprise and Edu users scheduled for the following week.
Pricing for the o3-pro model is set at $20 per million input tokens and $80 per million output tokens via OpenAI’s developer API. Input tokens refer to the data submitted into the system, while output tokens are the responses generated. OpenAI emphasises that the new model builds upon the capabilities of its predecessors by delivering enhanced reasoning skills. It is particularly suited for applications where precision takes precedence over speed, making it invaluable in areas requiring comprehensive analysis.
According to OpenAI, assessments by experts indicate that o3-pro consistently surpasses earlier models, including the original o3 model, across key criteria such as clarity, comprehensiveness, adherence to instructions, and accuracy. The ChatGPT developer claimed that the o3-pro model demonstrates notable strengths in areas such as science, education, programming, business, and writing support. OpenAI also highlighted the model’s performance in its “4/4 reliability” test, stating that it is deemed reliable if it successfully answers questions in all four trials.
Current limitations and ongoing technical improvements for o3-pro
Despite its capabilities, some limitations remain with the o3-pro model. Temporary chats have been disabled while technical issues are addressed. Furthermore, image generation and Canvas functionalities are not supported at present. Users requiring image generation capabilities will have to consider alternatives like GPT-4o or other OpenAI models designed for such tasks.
Last month, OpenAI introduced Codex, a cloud-based AI agent aimed at assisting software engineers. This tool is available to ChatGPT Pro, Team, and Enterprise users and seeks to streamline routine coding activities such as bug fixing and code reviews. Codex functions within secure sandbox environments that are preloaded with user repositories, ensuring security while rejecting malicious coding inquiries through updated policy safeguards.
Meanwhile, reports are emerging that CoreWeave is poised to benefit from a newly established partnership between Google’s cloud unit and OpenAI. According to Reuters, based on information from unnamed sources, CoreWeave is anticipated to provide computational resources to Google’s cloud division. Alphabet, Google’s parent company, plans to sell this computing power to OpenAI to meet the growing demand for AI applications like ChatGPT. Additionally, Google intends to offer some of its own computational resources to OpenAI, as noted by the sources who requested anonymity due to the sensitivity of the information. CoreWeave already has an existing $11.9bn contract with OpenAI, focusing on supplying crucial AI infrastructure and significantly boosting the computational capabilities for ChatGPT.
Read more: CoreWeave and OpenAI sign $11.9bn AI infrastructure deal
