OpenAI has released a new AI model called o3-mini, now available for use in ChatGPT with a new 'Reason' button. DeepSeek's AI chatbot has a similar button, accessible through the 'DeepThink (R1)' feature.

According to OpenAI, this model, first previewed in December 2024, is designed to be fast, cost-efficient, and highly skilled in STEM tasks like science, maths, and coding. It also aims to deliver better performance while keeping costs low and response times quick, says the company.

As per an official blog post by OpenAI, the o3-mini model comes with several features that make it useful for developers. It supports advanced tools like function calling, structured outputs, and developer messages, which can help developers integrate the model into their projects more easily. It also offers three levels of reasoning effort—low, medium, and high—so users can choose between faster responses or more detailed answers depending on their needs. Like previous models, the o3-mini supports streaming, allowing for real-time interactions.  

For ChatGPT users, o3-mini is now available to Plus, Team, and Pro subscribers, with Enterprise access set to arrive in February, says OpenAI. Free users can also try the model by selecting the 'Reason' option in the message composer. On the developer side, o3-mini is being rolled out to select users in API tiers 3-5, with support for the Chat Completions API, Assistants API, and Batch API.  

In terms of performance, OpenAI says that the o3-mini is faster and more accurate than its predecessor, the o1-mini. It delivers responses 24% quicker, with an average response time of 7.7 seconds, states an official post. OpenAI also says that expert external testers found that o3-mini produces clearer and more accurate answers, with 39% fewer major errors on difficult questions. The model apparently shines in STEM tasks, matching or even outperforming OpenAI's larger o1 model in areas like maths, coding, and science.  

OpenAI says they have also prioritised safety with o3-mini. The model uses a technique called deliberative alignment, where it reasons about safety guidelines before responding to user prompts. The company also says it has conducted thorough safety evaluations, similar to those used for its previous models, to ensure the model is safe and reliable.  

With this release, o3-mini replaces o1-mini in the ChatGPT model picker, offering higher rate limits and lower latency. For example, ChatGPT Plus and Team users will see their daily message limit increase from 50 to 150. OpenAI is also working on integrating search capabilities across its reasoning models, with o3-mini already supporting early prototypes of web-based search, states the official blog.

While o3-mini is optimised for STEM tasks, OpenAI's broader general knowledge model, o1, remains available for users who need more versatile capabilities, further adds OpenAI.