OpenAI’s o3-Mini: A Fast, Cheap, and Smart Unified Chat System to Answer Questions in Experimentation and Science

Some staffers claim that chat brings in the lion’s share of Openai’s revenue but o1 gets more attention from leadership. “Leadership doesn’t care about chat,” says a former employee who worked on (you guessed it) chat. There is no buzz for o1 since the code base wasn’t built for experimentation and it’s sexy. The former employee asked to remain anonymous, citing a nondisclosure agreement.

OpenAI spent a lot of time trying to fine-tuned the model that eventually became the advanced reasoning system called o1. The process of reinforcement learning trains models with a system of penalties and rewards. OpenAI pioneered reinforcement learning in order to create its advanced reasoning system, called R1. A former OpenAI researcher who is not authorized to speak publicly about the company says that reinforcement learning, applied to language models, works well for them.

Some inside OpenAI want the company to build a unified chat product, one model that can tell whether a question requires advanced reasoning. That has not happened so far. There is a drop-down menu in the chat where users can decide whether to use GPT-4o or o1 for most questions.

In response, OpenAI is preparing to launch a new model today, ahead of its originally planned schedule. The model will be available in both chat as well as an application. Sources say it has o1 level reasoning with 4o-level speed. In other words, it’s fast, cheap, smart, and designed to crush DeepSeek.

Originally announced as part of OpenAI’s 12 days of “ship-mas” in December, o3-mini is designed to match o1’s performance in math, coding, and science, while responding faster than the existing reasoning model. o3-mini should respond 24 percent quicker than o1-mini, and provide more accurate answers in the process. Much like o1-mini, this latest model will show how it worked out an answer, rather than just providing a response.