OpenAI launches o3-mini reasoning model to rival DeepSeek
What's the story
OpenAI has introduced its latest reasoning model, the o3-mini, into ChatGPT and its API services.
The new model was first announced as part of OpenAI's 12 days of "ship-mas" in December.
It aims to match the performance of o1 in math, coding, and science tasks while providing quicker responses.
As per OpenAI, the o3-mini should respond 24% faster than its predecessor, the o1-mini.
Enhanced performance
A step up in coding and reasoning tasks
Not only is the o3-mini model faster, but it's also more accurate than its predecessor. It shows its problem-solving process instead of just giving an answer.
Early benchmarks from OpenAI show that the o3-mini beats the o1 model in several coding and reasoning tasks.
The improved performance comes at a reduced cost and latency, making it a more efficient choice for developers.
Developer access
o3-mini now accessible through OpenAI's API services
Developers can now access the o3-mini model via several OpenAI's API services. These include the Chat Completions API, Assistants API, and Batch API.
A premium version of the model called o3-mini-high is also available for paid users. It provides more advanced responses at a slightly slower pace and is touted as the "best coding option in ChatGPT."
User access
Free users can now experience OpenAI's reasoning models
For the first time, free users of ChatGPT will get access to OpenAI's reasoning models.
This comes after Microsoft's recent move to make o1 available for all Copilot users and DeepSeek disrupted the AI landscape.
Users will be able to try out the o3-mini model for free in ChatGPT by choosing the "Reason" feature in the chat bar.
The rate limits for this free version will be similar to existing GPT-4o limits.
Model efficiency
o3-mini: A superior choice for STEM problems
The o3-mini model is particularly fine-tuned for STEM problems, such as programming, math, and science.
OpenAI claims that this new model matches the capabilities of the o1 family but works at a faster speed and lower cost.
External testers have preferred answers from the o3-mini over those from the o1-mini more than half of the time.
User availability
o3-mini to be available for ChatGPT Plus, Team, Pro users
The o3-mini model will also be available for ChatGPT Plus, Team, and Pro users globally.
OpenAI is tripling its message limits for Plus and Teams users to 150 messages per day. However, only Pro users willing to pay $200 a month will get unlimited access to the o3-mini model.
This new model can also be accessed through Microsoft Azure OpenAI Service, GitHub Copilot, and GitHub models as announced by Microsoft.