DeepSeek effect? OpenAI rolls out o3-Mini, it’s first reasoning model for free users

After previewing the o3-Mini reasoning model during the 12-day Ship-mas event in December, OpenAI is finally releasing the new model to both free and paid users. OpenAI's first-ever 'free' reasoning model comes as the company faces increasing competition from China's DeepSeek, whose R1 reasoning model has been freely available since its global debut in early January.

OpenAI says its new o3-Mini model has been optimised for tasks such as maths, coding and science, and matches the performance of o1 in most reasoning and assessment tests, while significantly outperforming o1-Mini with "more accurate and clearer answers".

o3-Mini has also been integrated with ChatGPT's web search system to provide "up-to-date answers with links to relevant web sources". However, the new model does not support vision capabilities, meaning users will not be able to share images and ask the model for answers.

What are reasoning models?

Reasoning models are large language models that are trained with reinforcement learning to perform complex tasks. Contrary to normal models like GPT-4, reasoning models are trained to produce a long internal chain of thought before responding which while making them a bit slower to respond also helps them excel in tasks like problem solving, coding, scientific reasoning, and multi-step planning.

While OpenAI's new o3-mini model almost matches the performance of DeepSeek R1, it costs $1.10 for 1 million input tokens, twice the price of its Chinese counterpart. Notably, since the recent release of DeepSeek's V3 and R1 models, there has been increased scrutiny of US AI companies.

Apart from shattering notions of American AI supremacy, DeepSeek also claimed in a paper that it spent just $5.6 million and used older Nividia H800 GPUs to train its V3 model, compared to the over $100 billion spent by OpenAI and others on similar models.

How to use o3-Mini?

o3-Mini is the first reasoning model available to free ChatGPT users. To access the model, just visit the official ChatGPT website or app and click on the new 'Reason' option available below the chat bar. While OpenAI did not explicitly state the usage limit for free users in its blog post, it is safe to assume that it would be similar to the levels of the GPT-4o model. Meanwhile, OpenAI has tripled the limit for ChatGPT Plus and Team users from 50 messages to 150 messages per day.

o3-Mini is available in three performance levels - low, medium and high - which developers can choose from according to their specific use cases. However, only ChatGPT Pro users will have 'unlimited access' to o3-mini and o3-mini high.