OpenAI’s o1-Pro: Advanced Reasoning And AI Agent Capabilities

OpenAI has launched a newer version of its o1 reasoning AI model which is called o1-pro, in its developer API. OpenAI claims that the o1-pro runs on more computational resources than its predecessor o1 to provide ‘superior responses with higher consistency.’ 

Right now, access to o1-pro is limited to a few selected developers willing to pay $5 for the OpenAI API service, and the price is known to be quite high. The pricing scheme for the same is $150/million tokens (roughly 750,000 words) into the model and $600/million tokens back out of the model. 

The price is two times that of OpenAI GPT-4.5 for input and ten times that of regular o1. OpenAI is banking on the improved capabilities of o1-pro to entice developers to cough up these substantial fees.

A representative of OpenAI who made a remark on the issues mentioned, “O1-pro in the API’ is a version of o1 that uses more computing power for better problem-solving and better answer generation.”

He further added, “At the behest of many in our developer community, we are beyond thrilled to launch it in the API in support of even more reliable responses.”

Nevertheless, the earliest comments about o1-pro have not been overwhelmingly favorable. Since launching in December, the software has been available for use by ChatGPT Pro subscribers. Among the complaints raised by users was the disastrous performance of the model when faced with Sudoku puzzles and an inability to understand very basic optical illusion jokes.

Earlier internal benchmarks run by OpenAI toward the end of last year indicated that o1-pro performed slightly better than standard o1 in code and not only this but also in mathematics tasks. However, the benchmarks also suggested it was more capable as well as reliable in solving these types of problems.