The company's most expensive AI model is OpenAI's o1-pro

OpenAI has introduced an enhanced version of its o1 “reasoning” artificial intelligence model, known as o1-pro, making it available to developers through its API. This upgraded model is designed to handle complex queries with greater computational power, promising more accurate and reliable responses compared to its predecessor, o1. However, access to o1-pro is currently limited to select developers who have spent a minimum of $5 on OpenAI API services.

What sets o1-pro apart is its ability to utilize more computing resources, allowing it to process information in a more sophisticated manner. OpenAI claims that this additional processing power results in improved problem-solving capabilities, making it a valuable tool for developers tackling intricate tasks. But there is a high price for this upgraded capacity for reasoning. OpenAI is charging developers a hefty $150 per million input tokens (equivalent to approximately 750,000 words) and an even steeper $600 per million output tokens generated by the model. To put this into perspective, o1-pro’s pricing is twice that of OpenAI’s GPT-4.5 for input tokens and a staggering ten times more expensive than the standard o1 model.

Despite its high price tag, OpenAI is confident that the model’s performance enhancements will justify the cost for developers who require more precise and reliable AI-driven solutions. In a conversation with TechCrunch, a spokesperson from OpenAI described o1-pro in the API as an advanced iteration of o1, leveraging increased computational resources to process information more effectively and deliver superior solutions to highly complex problems.” After receiving numerous requests from our developer community, we’re excited to bring it to the API, offering even more dependable responses.”

However, the initial reception to o1-pro has been somewhat mixed. The model has been available within OpenAI’s ChatGPT platform for ChatGPT Pro subscribers since December, and early user experiences have highlighted certain limitations. For example, some users reported that o1-pro struggled with solving Sudoku puzzles and failed to grasp simple optical illusion jokes, raising concerns about its reasoning capabilities.

Additionally, OpenAI’s internal testing from late last year indicated that while o1-pro exhibited marginally better performance in solving coding and mathematical problems compared to the original o1, the improvements were not as groundbreaking as some had hoped. The key advantage noted in these benchmarks was that o1-pro provided more consistent and reliable responses, which could be valuable in critical applications requiring high accuracy.

Despite these mixed early impressions, OpenAI remains optimistic about o1-pro’s potential to serve developers working on complex projects. Whether developers will be willing to pay the premium price for these incremental improvements remains to be seen, but OpenAI is betting that the demand for better reasoning AI models will make o1-pro a worthwhile investment.