Less than one week after DeepSeek’s “Sputnik moment,” OpenAI is releasing its o3-mini reasoning model to the public. This is the first OpenAI reasoning model that is not tied to a subscription—you can use it for free.
Reasoning models are far more accurate than typical LLMs. They utilize a “chain of thought” system to “think” before answering questions and self-correct mistakes. As a result, they’re the ideal option for difficult prompts, especially those that involve difficult math.
The new o3-mini reasoning model is specially designed for science, programming, engineering, and other math-heavy fields. And, per OpenAI, it can go toe-to-toe with the premiere o1 and o1-mini models. This is important because, at $1.10 per million input tokens, o3-mini is half the cost of o1-mini. It’s a leap in efficiency that would be recognized as a huge achievement, if not for DeepSeek.
At $1.10 per input token, o3-mini is two-times more expensive than DeepSeek’s reasoning model. And, if we’re to believe that DeepSeek’s model was trained for just $6 million, then o3-mini proves that last year’s AI leaders are far less efficient and far more wasteful than they should be.
I should also point out that o3-mini wasn’t expected this week. OpenAI appears to have fast-tracked its launch in response to DeepSeek’s meteoric rise. The fact that free users can access this model is also interesting, although we don’t know whether this particular point was influenced by DeepSeek or not—the DeepSeek model is notable for its open-source licensing.
To clarify, o3-mini is not open source, and free users do not have unlimited access. Pro users can tap into the model as much as they want, while Plus and Team users may simply enjoy the model’s reduced token cost.
To use o3-mini, press the “reason” button before submitting a question to ChatGPT. This option may not be immediately available in the ChatGPT desktop or mobile apps, but it’s live on the ChatGPT website.
Source: OpenAI
