The rumors were true. OpenAI just announced the o1 and o1-mini, the first AI models with logical reasoning capabilities. These models leverage technology known internally as Strawberry and are designed to solve complex tasks in fields like science, math, and programming.
Free users will also be able to use the template.
The new templates are available in paid versions of ChatGPT, and in a mini version. It will be added to the free version in the near future..
In our tests, the upcoming model update performs on par with PhD students on challenging benchmark tasks in physics, chemistry, and biology.
We also found that it excels at math and programming. In the International Mathematical Olympiad (IMO) qualifying test, GPT-4o solved only 13% of problems correctly, while the reasoning model scored 83%.
Countries OpenAI.
OpenAI has released the o1 in preview mode, So it has not yet integrated 100% of its features into ChatGPT.This means it can’t yet process images, nor can it search the web for information. Sam Altman’s team believes this new AI represents a quantum leap in capabilities, which is why they’ve decided to name it o1.
Startup Post some videos With examples showing how the new model solves complex problems and instructions, and even answers simple questions. that other models failed to answer..
OpenAI o1 answers a very difficult question for large language models. pic.twitter.com/5ZlQIOBWEd
- OpenAI (@OpenAI) September 12, 2024
ChatGPT gains reasoning capabilities through OpenAI's o1 and o1-mini
OpenAI says that o1 was trained using a custom dataset and a new optimization algorithm. Unlike GPT-4o and previous models, this AI doesn’t mimic your training patterns, but uses reinforcement learning to solve problems given to it on its own.
The company also said that the O1 is less vulnerable to jailbreaking, i.e. users trying to bypass its safeguards.
In one of the toughest jailbreak tests, GPT-4o scored 22 points (on a scale of 0 to 100), while the o1 preview model scored 84 points.
Say AI creators.
As for the o1-mini, it is a smaller and up to 80% cheaper version of the o1. According to OpenAI, the o1-mini particularly excels at programming tasks. An interesting fact shared is how long it takes your new models to provide a plain text response through ChatGPT:
- While GPT-4o responded in just 3 seconds, o1-mini responded in 9 seconds and o1 in 32 seconds. This extra time is the time the new AI spends thinking and thinking about its responses.
ChatGPT Plus and Team subscribers can now access o1 and o1-mini from the model selection menu. Next week, Enterprise and Edu plan users will also have access. As we said, the free version of ChatGPT will get o1-mini, Although it is not yet known when.
Read also:
“Coffee trailblazer. Social media ninja. Unapologetic web guru. Friendly music fan. Alcohol fanatic.”