About a month after its announcement, the o3-mini model has just officially appeared on ChatGPT, bringing many important improvements compared to previous models.
Starting January 31 (US time), global ChatGPT users can select the o3-mini model for queries. This model was introduced by OpenAI during the 12 Days of OpenAI event series that took place more than a month ago.
As a reasoning AI model, o3-mini has many improvements over o1-mini. Both are strong in mathematics and scientific programming, but o3 can answer about 24% faster.
According to OpenAI’s introduction, users participating in the test rated the o3-mini “for providing more accurate and clear answers, with stronger reasoning capabilities.”
Like the o1-mini, the o3-mini is a reasoning model, where the AI model “thinks”, reasoning step by step, when coming up with a final answer. The o3-mini has 3 different levels of reasoning depending on the use case: low, medium and high.
In the math test, the o3-mini’s medium and high reasoning levels outperformed the o1-mini at the same level, and the high-level o3-mini even outperformed the o1, a more powerful version of the o1-mini.
For PhD-level science questions, the o3-mini also beat the o1-mini at each of the low, medium, and high reasoning levels, but could not surpass the o1.
o3-mini will now replace o1-mini on ChatGPT for all user accounts. OpenAI did not reveal whether o1-mini will return in the future, only that o3-mini offers higher speeds and lower latency than the previous model. Free users can choose o3-mini.
ChatGPT Plus, Team, and Pro accounts can select a reasoning model before querying, including o3-mini. Enterprise accounts will also be able to access o3-mini in the same way next week. For free users, the option to switch models in the response section now includes o3-mini.
This is also the first time that free ChatGPT users can try OpenAI’s reasoning models, just days after Microsoft made o1 free to all Copilot users and DeepSeek shook the AI world.