• OpenAI’s o3-mini model is designed for reasoning-based tasks and is fine-tuned for STEM applications, including programming, math, and science.
  • It fact-checks itself before generating responses, making it more reliable than traditional AI models.
  • O3-mini is 63% cheaper than OpenAI’s o1-mini, priced at $0.55 per million input tokens and $4.40 per million output tokens.
  • It also offers faster performance, with responses delivered 24% quicker than o1-mini while maintaining similar capabilities.

OpenAI has officially launched o3-mini, its newest AI "reasoning" model, designed to deliver enhanced accuracy, faster performance, and affordability. This release follows OpenAI’s December preview of the o3 family and comes at a time when the company is navigating competitive and regulatory challenges.

The o3-mini model stands out from traditional large language models by fact-checking itself before generating responses, making it more reliable—especially for STEM applications like programming, math, and science. OpenAI claims o3-mini is on par with o1 and o1-mini but operates faster and at a lower cost.

External testing indicates that users preferred o3-mini’s answers over o1-mini’s more than 50% of the time, and the model made 39% fewer major mistakes on difficult real-world questions. Additionally, it delivers clearer responses while being 24% faster.

O3-mini is now available to all ChatGPT users, with premium subscribers getting higher query limits. It will roll out to ChatGPT Enterprise and ChatGPT Edu users next week. Developers can access it via OpenAI’s API, where they can adjust the model’s reasoning effort (low, medium, or high) based on accuracy and latency needs.

Pricing for o3-mini is 63% cheaper than o1-mini, costing $0.55 per million input tokens and $4.40 per million output tokens, making it competitive with DeepSeek’s R1 model.

Despite its strengths, o3-mini does not outperform all competitors across benchmarks. It surpasses R1 in some areas but lags in others, particularly in PhD-level science questions. Nevertheless, OpenAI asserts that o3-mini is safer than the o1 family, emphasizing its strong safety and jailbreak resistance.

Edited by Harshajit Sarmah