Introduction
OpenAI on 12th September 24 launched a new AI model named “Open AI o1” which they claim to have more versatility and more human-like reasoning. Open AI o1 can be more efficient than its predecessor models like ChatGPT 4o,3.5 turbo, etc as it can reason through complex tasks and is capable of solving harder problems. OpenAI has released it as the first of its series in ChatGPT and OpenAI API. It is just a preview and you can expect more updates coming frequently. If you are more interested in AI topics, please follow my website’s AI blog page
How Open AI o1 works
This new model is trained to do more thinking before giving the response which is different than previous models. Previous models used to provide quick responses that might provide incorrect information and this was termed as “Hallucination”. Here this model will work like a human in the real world which will think through the provide complex problems. Before responding it will keep on refining its thinking process by trying different strategies and recognizing its mistakes.
As per the official documentation, during their test, Open AI o1 performed similarly to a PhD student when given complex tasks related to Biology, Physics, and Chemistry It also performed well in Math and coding tasks as well.
Comparison with previous models
As Open AI o1 is an early model in this series, it doesn’t have advanced features like browsing through the web to provide information. Also, it doesn’t have features like file and image upload which are present in GPT- 4o. For these kinds of tasks ChatGPT 4o will be more capable as of now.
But for complex reasoning tasks, it is a significant change and provides a new level of AI capability. This is the reason that they have changed the naming convention and started with Open AI o1 instead of continuing with 3,3 turbo, 4o, and so on.
Safety Tests on Open AI o1
As per the official documentation, OpenAI has used a new safety training approach that will test its reasoning capability to follow safety and alignment guidelines.
On testing how well their model behaves when a User tries to bypass them (known as “jailbreaking”), the Open AI o1 model scored 84 as compared to ChatGPT 4o’s 22 score. This scorecard is available on OpenAI official page here. You can also read more about Open AI o1 reasoning here.
What are the Use cases for Open AI o1
If you’re dealing with challenging tasks in fields like science, coding, or mathematics, these enhanced reasoning capabilities can be a game-changer. The o1 model, for example, can help healthcare researchers annotate cell sequencing data, help physicists in generating complex mathematical formulas for quantum optics, and enable developers to create and execute multi-step workflows across various industries. It’s designed to handle complex tasks that require deep thinking and accuracy, making it an ideal tool for professionals in advanced fields.
Open AI o1-mini
The o1 series is outstanding at crafting and debugging intricate code with precision. To enhance the efficiency for developers, OpenAI is also introducing Open AI o1-mini—a more agile and economical reasoning model that’s especially adept at coding tasks. With its compact design, the Open AI o1-mini is 80% less expensive than the Open AI o1-preview, making it a potent yet affordable choice for applications that need reasoning capabilities without extensive general knowledge.
How to use Open AI o1
Starting from 12th September 24, ChatGPT Plus and Team users can access Open AI o1 models directly in ChatGPT. Users can manually select either o1-preview or o1-mini from the model picker. At the start, the weekly message limits will be 30 for o1-preview and 50 for o1-mini. The OpenAI team is expected to raise these limits and enable ChatGPT to automatically select the most suitable model based on the prompt.
While Upgrading Plan
For Existing ChatGPT plus users
Developers eligible for API usage tier 5 can now begin prototyping with both models in the API, with a rate limit of 20 requests per minute. Again OpenAI team is expected to raise these limits following further testing. Currently, the API for these models does not support function calling, streaming, system messages, or other features. For more information on getting started, please refer to the API documentation.
Good News for Free members !!. OpenAI team is also planning to bring o1-mini access to all ChatGPT Free users in the upcoming days
What’s in the Future
This is an early preview of the reasoning models in both ChatGPT and the API. Along with updates to the models, and features like browsing, file, and image uploads are expected to be added with upcoming updates.
Regarding future releases, OpenAI is also planning to keep developing and releasing new models in our GPT series, alongside the new Open AI o1 series.