OpenAI today released o3-mini, a new, more cost efficient model that's available for both ChatGPT and the OpenAI API starting today. OpenAI previewed o3-mini in December, but it is ready at a time when OpenAI is aiming to keep up with Chinese company DeepSeek.
According to OpenAI, o3-mini "advances the boundaries" of what's possible with smaller, more efficient models. o3-mini is designed for STEM capabilities, and OpenAI says that it has particular strength in science, math, and coding. o3-mini is search capable, and it can locate up-to-date answers with links to relevant web sources.
o3-mini is OpenAI's first small reasoning model that supports features like function calling, Structured Outputs, and developer messages. There are three reasoning effort options, including low, medium, and high, to optimize for different use cases that prioritize processing power for complex challenges or speed.
OpenAI is keeping o1 as its standard general knowledge reasoning model, but o3-mini is available for technical domains that require precision and speed. When used with ChatGPT, o3-mini uses medium reasoning effort for a balance between speed and accuracy, but o3-mini high is an option in the model picker.
ChatGPT Plus, Team, and Pro users are able to use o3-mini starting today, and it will replace o1-mini in the model picker. The rate limit for Plus and Team users is now 150 messages per day, up from 50.
o3-mini is the first reasoning model that OpenAI is making available to free users. Those with the free plan can try o3-mini by selecting the "Reason" option in the message composer.
/https://forums.macrumors.com/attachments/2477855/)
