OpenAI has introduced o1, its reasoning AI model, to select developers starting Tuesday. This rollout targets those in the 'level 5' usage category, requiring a minimum spend of $1,000 and an account older than 30 days.
Replacing the previous o1-preview model, o1 offers enhanced fact-checking capabilities, helping to mitigate common pitfalls of AI models. However, this improvement comes at a cost, as the model typically requires more processing time and resources, with OpenAI charging $15 for approximately 750,000 words of analysis and $60 for generated content, significantly higher than its predecessor, GPT-4o.
The new o1 model is considerably more customizable, featuring function calling, developer messaging for tone and style instructions, and image analysis. It also introduces an API parameter, 'reasoning_effort,' which allows developers to control the model’s processing time before responding.
OpenAI describes this version of o1 as a 'new post-trained' model, promising more comprehensive and accurate responses, particularly in programming and business inquiries. The company is gradually expanding access while improving usage limits.
In related news, OpenAI announced new versions of its mini-models, GPT-4o and GPT-4o, within its Realtime API, which remains in beta but has added features such as off-band simultaneous responses. The API now supports WebRTC, facilitating real-time voice applications across various devices.
OpenAI's integration of WebRTC aims to ensure seamless interactions under real-world conditions, managing audio encoding, streaming, noise suppression, and congestion control. Additionally, the company has refined its fine-tuning preferences API and launched a beta version for official SDKs in Go and Java.