o3-mini and o3-mini (high) will be released today.
Regular users will also get o3-mini, and plus users will be able to use o3-mini (high).
o3-mini (high) is about 200 points higher than o1 on Codeforce, faster than o1, and performs better in coding and mathematics, but the cost is still at the level of o1-mini.
Plus users can use o3-mini 100 times a day. However, the usage limit of o3-mini (high) needs to be further confirmed.
Some netizens said, yes, R1 is so popular that Openai can’t hold back:
And earlier, the Alibaba Qwen team released qwen2.5-max on New Year’s Eve. During the Spring Festival, everyone still needs to roll, hahaha…
In fact, as early as the Christmas live broadcast, Openai announced that o3 mini will be available in early 2025:
We still need to talk about what o3 and o3-mini are?
o3: A cutting-edge inference model that excels in coding, mathematics, and even AGI-oriented benchmark tests. It sets a new benchmark for intelligence and problem solving.
o3-mini: A cost-effective version of o3 that provides superior performance at a very low cost and speed.
These models have taken inference to a whole new level, making breakthroughs in complex tasks possible that require in-depth understanding and logic.
o3 brings three major breakthroughs.
Programming ability: 71.7% accuracy in practical programming, 20% higher than o1. 2727 points on Codeforces, already surpassing human level.
Math level: nearly 97% accuracy in the US Mathematics Olympiad qualifying round. Even the most difficult Epic AI frontier math problems can get 25% results.
The most amazing thing is the Arc AGI test: 87.5%, surpassing humans for the first time on this extremely difficult benchmark test.
Why is the o3-mini a disruptive innovation? The o3-mini brings two changes.
Adaptive thinking: the depth of reasoning can be adjusted according to the difficulty of the task, with three modes to choose from: low, medium, and high.
This makes the AI more closely aligned with real-world usage scenarios.
Cost-effectiveness breakthrough: lower cost than the o1-mini, faster response, and better results.
However, netizens lament that o3 high consumes $1,000 per task:
In addition, there are indeed too many models available, and we have yet to confirm how to switch between them.
