The new model comes at a time when the AI industry is facing uncertainty.
The newest machine learning model from OpenAI has been released. The company released o3-mini on Friday, and it is now accessible for you to try. Additionally, OpenAI is making one of its "reasoning" models available to users of ChatGPT for free for the first time. If you would want to give it a shot on your own, click the "Reason" button that is located beneath the message composer to begin.
OpenAI states that o3-mini is quicker and more precise than the previous version, o1-mini. The company discovered that o3-mini was 24 percent faster than o1 at delivering a response during A/B testing. Additionally, when the new model is set to its "medium" reasoning effort, it can perform similarly to the more expensive o1 system in some arithmetic, coding, and scientific benchmarks. O3-mini, like OpenAI's other reasoning models, will show you how it arrived at an answer instead of merely replying to a question. It is worth mentioning that the model is compatible with ChatGPT Search right away, allowing it to search the internet for the most up-to-date information and helpful links. According to OpenAI, the company is now working on incorporating search functionality into all of its reasoning models.
The release of OpenAI o3-mini is another milestone in OpenAI's mission to expand the limits of affordable intelligence. OpenAI stated, "By optimizing reasoning for STEM domains while keeping costs low, we’re making high-quality AI even more accessible." "This approach maintains our history of lowering the cost of intelligence—reducing per-token price by 95% since the launch of GPT-4—while still preserving top-tier reasoning skills. We are dedicated to staying at the forefront of artificial intelligence as its use becomes more widespread. We will continue to develop models that maintain a balance between intelligence, efficiency, and safety on a large scale.
As a result of today's announcement, o3-mini will take the position of o1-mini in the model picker. Additionally, OpenAI is increasing the rate limit for Plus and Team ChatGPT customers from 50 messages per day with o1-mini to 150 messages per day for o3-mini. The Pro tier, which OpenAI recently announced for $200 a month, provides unrestricted access to the new system.
At the end of last year, OpenAI gave its first glimpse of o3 and o3-mini. During this preview, CEO Sam Altman stated that o3-mini would be available "around the end of January." On January 17, Altman provided a more specific schedule when he posted on X that OpenAI was "planning to ship in a couple of weeks."
Now that it has here, we can confidently declare that o3-mini comes with a sense of urgency. On January 20, the same day that Altman was attending Donald Trump's inauguration, China's DeepSeek surreptitiously unveiled its R1 chain-of-thought model. After going viral, the company's chatbot became the most downloaded free software in the US software Store by January 27, surpassing ChatGPT. DeepSeek's rapid breakthrough wiped off $1 trillion in stock market value and almost definitely caught OpenAI off guard.
Following last week, OpenAI announced that it was collaborating with Microsoft to find two accounts that the company believes may have distilled its models. The process of distillation involves converting the knowledge of a more powerful AI system to a smaller, more efficient AI system. Distillation is a practice that is not considered controversial. DeepSeek has trained its smaller algorithms by using distillation on its own R1 model. In reality, OpenAI's terms of service allow for distillation as long as users do not train competitive models on the outputs of the company's AI.
DeepSeek was not specifically mentioned by OpenAI. A spokesman for OpenAI recently told The Guardian, "We know [China]-based companies — and others — are constantly trying to distill the models of leading US AI companies." On the other hand, David Sacks, who is an advisor to President Trump on artificial intelligence, was more straightforward in his comments, stating that there was "substantial evidence" that DeepSeek had "distilled the knowledge out of OpenAI's models."
