Last week, OpenAI introduced a new series of AI models designed to spend more time thinking before they respond. The new OpenAI o1 series models can reason through complex tasks and solve harder problems than previous OpenAI models in science, coding, and math. For example, o1 models scored in the 89th percentile on competitive programming questions (Codeforces) and placed among the top 500 students in the US in a qualifier for the USA Math Olympiad (AIME).
The OpenAI o1 series includes two reasoning models: o1-preview, an early preview of the o1 model, and o1-mini, a faster and cheaper version of o1. It is important to note that o1 models only offer significant advancements in reasoning, so they cannot replace GPT-4o in all use cases. o1 models do not support image inputs, function calling, or offer consistently fast response times. So, the GPT-4o and GPT-4o mini models will continue to be the choice for most developers. Developers who need deep reasoning and can accommodate longer response times can use these new o1 models.
Early this week, OpenAI increased the rate limit of o1 series model APIs for developers on tier 5. o1-preview supported 100 requests per minute, while o1-mini supported 250 requests per minute. Today, OpenAI is further increasing the rate limits of these APIs. The o1-preview API now supports 500 requests per minute, while the o1-mini API now supports 1000 requests per minute. The OpenAI team is working further to increase the rate limits and expand the API access to more developer tiers.
For general consumers, OpenAI increased the message limits early this week. For ChatGPT Plus and Team users, OpenAI increased rate limits for o1-mini by 7x, from 50 messages per week to 50 messages per day. Since the o1-preview model is more expensive to serve, OpenAI increased the rate limit from 30 messages per week to 50 messages per week. We can expect OpenAI to further increase message limits for general consumers in the coming months as they optimize these new models for cost.
With continuous improvements and expanded access, the o1 series models hold great promise for various applications, particularly those demanding advanced reasoning abilities.