ChatGPT o1 ‘Strawberry’ is here and it is the first ‘Reasoning’ model by OpenAI

Everybody was talking about OpenAI's next announcement for Strawberry or Strawberry/Q and the AI giant once again changed the model name by releasing ChatGPT o1 and ChatGPT o1 mini with the most awaited "Reasoning" capabilities.

Image from OpenAI

Yes,

OpenAI introduced a new series of AI models designed for advanced reasoning (just like a human with a PhD), this new model is capable of solving harder problems in science, coding, and math by spending more time thinking before responding to the prompt.

The first model in this series "ChatGPT o1" was released on September 12, 2024 (yesterday), as a preview in ChatGPT and available through the APIs.

ChatGPT o1 model is specifically designed to act like a human and here's a list of some of its capabilities:

• Training Focus: ChatGPT o1 models are trained to refine their thinking process, try different strategies, and recognize mistakes, similar to human reasoning capabilities.
• Performance: The next model update performs at a level comparable to PhD students in challenging tasks across physics, chemistry, and biology. It excels in math and coding, significantly outperforming previous models.

In an official announcement, OpenAI shared that:

In an International Mathematics Olympiad (IMO) qualifying exam, the reasoning model scored 83%, while GPT-4o scored 13%.

This means this ChatGPT o1 model is far better than ChatGPT 4o and ChatGPT Plus.

• Tip: You can access this model from this URL by going to this web-page: https://chatgpt.com/?model=o1-preview

There are so many safety features in ChatGPT o1 and for all the good reasons. Here's what I have read and understood about the safety features of this new model from the official statements:

• Enhanced Safety Training: The models' reasoning capabilities improve adherence to safety and alignment guidelines.
• Collaborations: OpenAI has formalized agreements with U.S. and U.K. AI Safety Institutes, granting early access to a research version of the model to advance AI safety.

The AI giant claimed:

The o1-preview model scored 84 out of 100 in a challenging jailbreaking test, compared to GPT-4o's score of 22.

Image from OpenAI

Forcing me to use o1 rather than 4o, but wait, there's OpenAI o1-mini too.

And here's why o1 mini is also a great model:

• Cost-Effective Option: OpenAI o1-mini is a faster, cheaper version, particularly effective at coding, and 80% cheaper than the o1-preview model.
• Availability: It is designed for applications requiring reasoning but not broad-world knowledge.

So, if you are a student or a developer, you don't need to pay for the o1 model, you can go for the o1 mini to save some dollars every month.

While these new models are available to everyone on a Plus account, OpenAI plans to expand access to ChatGPT Free users and add more features like browsing and file uploading.

Meaning if you can't pay right now to access ChatGPT o1 (ChatGPT Strawberry) models, you can wait for a few months to access them as they will finally be there for free users.

That's all for today.