OpenAI has launched a extra highly effective model of its o1 “reasoning” AI mannequin, o1-pro, in its developer API.
In line with OpenAI, o1-pro makes use of extra computing than o1 to offer “persistently higher responses.” Presently, it’s solely out there to pick out builders — those that’ve spent no less than $5 on OpenAI API companies — and it’s expensive. Very expensive.
OpenAI is charging $150 per million tokens (~750,000 phrases) fed into the mannequin and $600 per million tokens generated by the mannequin. That’s twice the worth of OpenAI’s GPT-4.5 for enter and 10x the worth of normal o1.
OpenAI is betting that o1-pro’s improved efficiency will persuade builders to pay these princely sums.
“O1-pro within the API is a model of o1 that makes use of extra computing to assume tougher and supply even higher solutions to the toughest issues,” an OpenAI spokesperson advised TechCrunch. “After getting many requests from our developer neighborhood, we’re excited to convey it to the API to supply much more dependable responses.”
But early impressions of o1-pro, which has been out there in OpenAI’s AI-powered chatbot platform ChatGPT for ChatGPT Professional subscribers since December, weren’t extremely constructive. The mannequin struggled with Sudoku puzzles, customers discovered, and was tripped up by easy optical phantasm jokes.
Moreover, sure OpenAI inner benchmarks from late final yr confirmed that o1-pro carried out solely barely higher than the usual o1 on coding and math issues. It did reply these issues extra reliably, nevertheless, the benchmarks discovered.
