OpenAI launches o3-mini: the new AI model for advanced reasoning | OpenAI Login | OpenAI API | ChatGPT 4 | Turtles AI

OpenAI launches o3-mini: the new AI model for advanced reasoning
OpenAI Introduces o3-mini: A New AI Model That’s Faster, More Accurate, and Affordable
Isabella V31 January 2025

 

OpenAI introduced o3-mini, an AI model optimized for STEM reasoning, with high performance in math, programming, and science. It provides faster, more reliable answers, while costing less than previous models in the o1 family. Available in ChatGPT and API, it supports a variable level of reasoning depth to balance speed and accuracy.

Key Points:

  • Efficiency and accuracy: o3-mini reduces errors and improves clarity of answers.
  • STEM optimization: especially effective in math, programming, and science.
  • Flexible access: available for free users and subscribers, with exclusive benefits for premium plans.
  • Low cost: 63% cheaper than o1-mini and competitive with DeepSeek’s R1 model.

OpenAI has officially released o3-mini, the latest model in its advanced reasoning line. Previewed last December, the new system is designed to offer an ideal compromise between computational power, speed and affordability. The company emphasizes that o3-mini represents a significant step forward in improving reasoning capabilities, especially for applications related to STEM disciplines, including mathematics, programming and science. Unlike traditional language models, the new system thoroughly checks the data before formulating a response, minimizing errors and ensuring greater reliability.

One of the most innovative aspects of o3-mini is its ability to operate at different levels of reasoning effort, adjustable based on the user’s needs. The default mode in ChatGPT uses a medium level, which balances speed and accuracy, but users of premium plans can select the o3-mini-high version, which maximizes the intelligence of the model in exchange for slightly longer response times. This makes it particularly competitive with alternatives such as DeepSeek R1, against which it competes directly in terms of performance and cost. In internal testing, o3-mini outperformed R1 in certain benchmarks, such as AIME 2024 and SWE-bench Verified, but with a high reasoning effort. However, in less complex contexts, the Chinese model maintains a slight advantage.

On the economic side, OpenAI has made o3-mini more accessible than previous models in the o1 line, with a cost of $0.55 per million input tokens and $4.40 per million output tokens. This represents a 63% reduction from o1-mini, making it a more cost-effective solution for both independent developers and enterprises seeking AI specialized in advanced computation and analytics. Developers can access the model via API and select the level of reasoning depth based on their needs, an option that improves the scalability of AI-based applications.

The new model is available in ChatGPT for all users, with advanced features available for Plus, Team, and Pro subscribers. Free plan users can try out o3-mini by selecting the “Reason” button in the message composer or by regenerating a reply. For premium subscribers, the daily speed limit has been increased to 150 queries, while Pro users have unlimited access. OpenAI also confirmed that o3-mini will be integrated into ChatGPT Enterprise and ChatGPT Edu within a week.

Another key aspect is compatibility with web search, an experimental feature that allows o3-mini to integrate up-to-date information into replies, including links to verified sources. This tool is still in development, but it represents an early attempt to combine AI and real-time search. However, OpenAI specifies that o3-mini does not support image analysis, a feature that is still the preserve of o1.

On the security side, o3-mini implements an advanced methodology called “deliberative alignment,” which allows it to consider OpenAI’s security policies before formulating a response. In internal testing, the model has demonstrated superior resistance to jailbreak and content manipulation attempts compared to GPT-4o, providing greater robustness against misuse.

With the launch of o3-mini, OpenAI continues its goal of making AI more accessible and efficient, improving the balance between performance, cost, and security.