OpenAI has launched o3-pro, its most capable AI model yet, enhancing its o3 reasoning model introduced earlier this year. The new model is designed to process problems sequentially, step-by-step.
Reasoning models like o3 and o3-pro are intended to improve performance in complex domains such as physics, mathematics, and programming by handling problems in a sequential manner. This capability is expected to yield better outcomes in these fields.
Starting Tuesday, o3-pro is available to users of ChatGPT Pro and Team plans, replacing the previous o1-pro model. OpenAI stated that Enterprise and Edu users will gain access to o3-pro in the subsequent week, ensuring a phased rollout to different user categories.
In addition to its availability within ChatGPT, the o3-pro model is now live on OpenAI’s developer API. The pricing for the API is set at $20 per million input tokens and $80 per million output tokens. To put this into perspective, OpenAI clarifies that one million input tokens are roughly equivalent to 750,000 words, providing a scale for the pricing.
OpenAI highlighted the model’s performance improvements, stating in a changelog, “In expert evaluations, reviewers consistently prefer o3-pro over o3 in every tested category and especially in key domains like science, education, programming, business, and writing help.” The company also noted that reviewers rated o3-pro higher for “clarity, comprehensiveness, instruction-following, and accuracy,” underscoring its enhanced capabilities.
O3-pro integrates with various tools, enabling capabilities such as web browsing, file analysis, understanding visual inputs, using Python, and personalizing responses through memory features. This integration is aimed at making the model more versatile and useful across different applications. However, OpenAI acknowledged a drawback: o3-pro responses typically take longer to generate compared to o1-pro, which could impact user experience.
The new model also has certain limitations upon release. Temporary chat sessions with o3-pro are currently disabled in ChatGPT due to an ongoing technical issue. Additionally, o3-pro does not support image generation, nor does it integrate with Canvas, OpenAI’s AI-powered workspace feature. These limitations are expected to be addressed in future updates.
Despite these interim limitations, OpenAI presented internal testing results that indicate strong performance on standard AI benchmarks. According to the company’s data, o3-pro performed better than Google’s Gemini 2.5 Pro on the AIME 2024 math skills test. Furthermore, o3-pro surpassed Anthropic’s recently released Claude 4 Opus on GPQA Diamond, a benchmark testing knowledge at the PhD level in science, demonstrating its competitive edge.
OpenAI specified that as o3-pro shares the same core model as o3, comprehensive safety details can be found in the o3 system card, providing users with a reference point for understanding the safety aspects of the new model.
