OpenAI o1-Pro now in API: a more powerful but costly reasoning LLM | OpenAI Playground | OpenAI API | OpenAI italiano | Turtles AI

OpenAI has introduced o1-pro, an advanced AI model designed to improve reasoning capabilities. Available through its API to select developers, o1-pro offers better performance than previous models, but at a significantly higher cost. Early evaluations show improvements in specific areas, although some limitations remain.

Key Points:

• Enhanced performance: o1-pro uses increased processing power to provide more accurate and reliable responses.
• Selective access: Currently restricted to developers who have spent at least $5 on OpenAI API services.
• High cost: Fees of $150 per million input tokens and $600 per million generated tokens, higher than previous models.
• Mixed evaluations: Although o1-pro showed improvements, it struggled with some complex tasks.

OpenAI recently expanded its AI model portfolio by introducing o1-pro, an enhanced version of its o1 “reasoning” model. This new model has been integrated into the developer API, offering higher performance through increased processing power. According to an OpenAI spokesperson, o1-pro is designed to “think deeper and provide even better answers to the toughest problems,” responding to the developer community’s calls for more reliable solutions.

Access to o1-pro is currently limited to select developers, specifically those who have spent at least $5 on OpenAI API services. This strategy is intended to ensure that the model is used by professionals with experience integrating advanced AI solutions. However, economics are a significant consideration: OpenAI charges $150 per million tokens of input and $600 per million tokens generated by the model. These costs are double those associated with GPT-4.5 for input and ten times higher than those of the standard o1 model. The decision to set these high prices reflects OpenAI’s confidence in the added value that o1-pro offers, with the expectation that its advanced performance will justify the investment for developers.

Despite high expectations, first impressions of o1-pro have been mixed. Available on the ChatGPT platform for ChatGPT Pro subscribers since December, the model has shown some difficulty with specific tasks. For example, some users reported that o1-pro struggled with puzzles like Sudoku and showed uncertainty when faced with simple optical illusions. Additionally, OpenAI’s internal benchmarks late last year indicated that while o1-pro performed slightly better than the standard o1 model on coding and math problems, the differences were not substantial. However, it was noted that o1-pro provided answers more consistently and reliably, suggesting improved performance stability.

OpenAI’s decision to launch o1-pro comes amid a rapidly evolving AI landscape, with demand for more powerful and versatile models growing. The availability of o1-pro through the developer API represents an opportunity to integrate advanced reasoning capabilities into real-world applications, although practical implications and widespread adoption remain to be assessed. 

o1-pro represents a significant step in the evolution of OpenAI’s AI models, offering advanced capabilities coupled with high costs and operational challenges that require further evaluation by the developer community.