OpenAI has launched o3-mini, its most cost-efficient AI model today. This model, designed as a small reasoning tool, is available on ChatGPT and through OpenAI’s API and is expected to excel in tasks related to science, mathematics, and coding. The company introduced almost a week later than DeepSeek, the Chinese model everybody’s talking about recently as they launched DeepSeek R1, which is almost 100x cheaper via API and free on its chat app.
As the first of its kind in the company’s “small reasoning model” category, o3-mini uses a train-of-thought process to improve task completion accuracy. Testing showed that o3-mini outperforms its predecessor, o1-mini, on multiple math and coding benchmarks, and sometimes even exceeds the full o1 model’s performance. Users will have control over the reasoning effort exerted by o3-mini, which can aid in cost management for developers working on applications that do not necessitate complete reasoning.
Subscribers to ChatGPT’s $200 monthly Pro tier will have unlimited access to o3-mini, while those at the $20 Plus tier will be able to use the model for up to 150 messages per day. Free users will also be able to experiment with o3-mini, though the duration of this access is not specified. Developers utilizing OpenAI’s API for new applications with o3-mini will incur costs of $1.10 per million input tokens and $4.40 per million output tokens.
OpenAI strengthens government partnerships
The release of o3-mini comes during a significant week for OpenAI, which includes the announcement of ChatGPT Gov, a model tailored for U.S. government agency use. ChatGPT Gov facilitates agency deployment in their own Microsoft Azure cloud environment, allowing agencies to handle security and compliance requirements more efficiently. OpenAI believes this service will streamline the analysis of non-public sensitive data.
Since 2024, over 90,000 users from more than 3,500 U.S. government entities have engaged with ChatGPT, sending more than 18 million messages for everyday task support. Notably, the Air Force Research Laboratory uses the tool for administrative tasks, and a pilot program in Pennsylvania has reportedly led to a deduction of approximately 105 minutes per day in routine work.
Applications of o3-mini in research and security
OpenAI has also announced a partnership with the National Laboratories to facilitate research breakthroughs using o3-mini. The agreement involves deploying current and future flagship AI models on Venado, a supercomputer in Los Alamos, developed in collaboration with Nvidia. This facility is intended to drive advancements in materials science, renewable energy, and astrophysics, providing a shared resource for Los Alamos, Lawrence Livermore, and Sandia National Labs.
Researchers will explore the model’s capabilities in identifying new disease treatments, enhancing national security threat detection, and optimizing the utilization of natural resources. The model will also support Los Alamos’ nuclear security program, with specific use cases considered on a case-by-case basis, in coordination with government officials and OpenAI researchers who hold security clearances.
Katrina Mulligan, OpenAI’s lead for national security policy and partnerships, noted that she joined the company to influence significant national security decisions. The partnership with the National Labs, as announced on January 30, aligns with her goal of contributing to game-changing decisions in scientific fields.
The model launch reflects OpenAI’s strategic response to the competitive landscape of AI development, particularly in light of the advancements made by rival company DeepSeek. Although let’s not forget, Sam Altman, OpenAI CEO first congratulated DeepSeek, and two days later the company accused them of stealing their model.
