OpenAI launched its new flagship AI models, o3 and o3-mini, on February 1, 2025, offering users advanced reasoning features for the first time.
The o3-mini model has been enhanced with an updated 'chain of thought' (CoT) feature, which improves transparency by displaying the reasoning process behind its answers.
To utilize the CoT feature, users can select the o3-mini model for their prompts, with free-tier users activating it by clicking 'Reason' or regenerating responses.
In a demonstration, the model effectively showcased its reasoning steps when answering questions, aiding users in verifying accuracy and learning problem-solving techniques.
The o3-mini model is capable of retrieving real-time data from the web, providing answers with source links and processing questions in approximately six seconds.
Paid subscribers can access a 'high reasoning' version of o3-mini, which utilizes more computational power for handling complex queries.
Premium users benefit from more detailed insights into the model's thought process through the 'high reasoning' mode.
While OpenAI's approach aims to simplify user understanding and mitigate security vulnerabilities, some critics argue that summarized CoT may obscure errors and complicate debugging.
The updated CoT in o3-mini summarizes reasoning steps rather than displaying them in real-time, leading to criticism from AI enthusiasts who prefer raw CoT for better troubleshooting.
In comparison, DeepSeek's R1 model fully reveals its thought process, which some AI researchers prefer for enhanced transparency and user experience.
An example of the model's response to a humorous question illustrates its capability to understand user emotions and context.
OpenAI has chosen not to display complete reasoning steps for o3-mini due to competitive concerns, opting instead for condensed summaries that have occasionally shown inaccuracies.
