OpenAI Expands Access to Premium o1 Reasoning Model in API
OpenAI releases a highly customizable version of its o1 reasoning model to its API for top-tier developers. The update brings new controls and improved accuracy, though it comes at a significantly higher price point than standard models.
OpenAI releases its o1 reasoning model to the API for developers in its tier 5 usage category, which requires spending at least $1,000 and having an account older than 30 days. This new version replaces the previous o1-preview and offers significantly more customization through features like function calling, developer messages, and image analysis. A new "reasoning_effort" parameter also gives developers direct control over how long the model processes a query before generating a response.
The o1 model sets itself apart from standard AI by effectively fact-checking its own logic, which reduces common errors and incorrect refusals, especially in programming and business contexts. However, this advanced self-correction requires substantial computing power, making it much more expensive to operate. OpenAI charges $15 per 750,000 words analyzed and $60 per 750,000 words generated, pricing it at six times the cost of the standard GPT-4o model.
Alongside the o1 release, OpenAI introduces updated versions of its GPT-4o and GPT-4o mini models specifically for the Realtime API. These new real-time voice models promise improved data efficiency and greater reliability while remaining cheaper to use. OpenAI plans to expand API access to the o1 model gradually as it works to increase rate limits for more usage tiers.