OpenAI Just Released o3-mini, Its Most Cost-Efficient Model Yet

OpenAI just released o3-mini, a miniature version of its upcoming flagship AI model. The new model is the company’s first “small reasoning model,” capable of using a train-of-thought process to complete tasks more accurately. The model’s launch, now available both on ChatGPT and through OpenAI’s API, caps off a week that also saw the company strengthen its ties with the United States government in the form of announcements about ChatGPT Gov and a partnership with the U.S. National Laboratories. In a blog post today, OpenAI shared that it anticipates o3-mini will be particularly useful for tasks involving science, math, and coding. The company’s testing indicates that o3-mini outperforms its predecessor, o1-mini, across several math and coding benchmarks, and in some aspects even outperforms the full o1 model. Like o1, users will be able to determine how much effort o3-mini puts into its reasoning, which could help developers save money when building applications that don’t require full effort. Subscribers at ChatGPT’s $200 per month Pro tier will get unlimited access to o3-mini, while those who pay $20 for ChatGPT’s Plus tier will be allowed 150 messages to o3-mini per day. Free users will also get a chance to try the model, but it’s unclear for how long. Developers who want to use OpenAI’s API to create new applications with o3-mini will pay $1.10 per one million input tokens and $4.40 per million outputs tokens. (Tokens are grammar elements that have been converted into data that can be processed by an AI model.) The model’s launch comes just as OpenAI is wrapping up a big week. On Tuesday, the AI market leader announced ChatGPT Gov, a version of the LLM that’s been tailored “to provide U.S. government agencies with an additional way to access OpenAI’s frontier models.” In a blog post announcing ChatGPT Gov, OpenAI wrote that government agencies will be able to deploy the system in their own Microsoft Azure cloud environment. “Self-hosting ChatGPT Gov,” according to OpenAI, “enables agencies to more easily manage their own security, privacy, and compliance requirements, such as stringent cybersecurity frameworks.” The company added that it anticipates this new service will make it easier for government agencies to approve the analysis of “non-public sensitive data” by ChatGPT. ChatGPT Gov has a similar operating system to ChatGPT Enterprise, OpenAI’s business-focused product. OpenAI also shared that since 2024, “more than 90,000 users across more than 3,500 US federal, state, and local government agencies have sent over 18 million messages on ChatGPT to support their day-to-day work.” The Air Force Research Laboratory uses the tool for basic administrative support, and the Commonwealth of Pennsylvania is taking part in a pilot program with ChatGPT that OpenAI claims has reduced the time spent on routine tasks by “approximately 105 minutes per day on the days they used it.” ChatGPT has also been used for some time to enhance research at Los Alamos National Laboratory in New Mexico, the birthplace of the atomic bomb. But a major new deal means OpenAI will soon have an even larger presence there. The company announced on Thursday that it has agreed to deploy current and future flagship AI models on Venado, a supercomputer in Los Alamos built in collaboration with Nvidia. According to OpenAI’s blog post announcing the deal, the computer was designed to “drive scientific breakthroughs in materials science, renewable energy, astrophysics, and more,” and it will be a shared resource for researchers at Los Alamos, Lawrence Livermore, and Sandia National Labs. As for how the models will be used, OpenAI says researchers will probe the technology for its potential to identify new approaches for treating and preventing diseases, improve detection of national security threats, and unlock “the full potential of natural resources.” The models will also be used to support Los Alamos’ nuclear security program, but use cases will be carefully decided on an individual basis in consultation with government officials and OpenAI researchers with security clearances. In a post on Linkedin, OpenAI national security policy and partnerships lead Katrina Mulligan said that she joined OpenAI “because I believed that some of the most consequential national security decisions of the decade would be made at companies like this and I wanted a seat at that table. Today’s announcement of our partnership with the National Labs to advance the future of science is exactly the kind of game-changing decision I wanted to have a role in making.” BY BEN SHERRY @BENLUCASSHERRY