OpenAI introduces its new 'OpenAI o1' model with advanced reasoning abilities
OpenAI has introduced its latest AI model, "OpenAI o1", aimed at handling more intricate tasks than previous models like GPT-4 or the more recent GPT-4o Mini. The new naming convention marks a shift from "ChatGPT" or "GPT" to "OpenAI o1." This model is designed to spend more time processing before responding, making it suitable for complex subjects such as physics, chemistry, and biology.
In testing, OpenAI o1 performed on par with PhD students in challenging benchmark tasks and excelled in math and coding. It solved 83% of problems in an International Mathematics Olympiad qualifier, a significant improvement over GPT-4's 13%, and placed in the 89th percentile in Codeforces coding competitions. However, it lacks the ability to browse the web or analyze uploaded files, capabilities that are present in ChatGPT LLMs.
OpenAI also launched OpenAI o1-mini, a streamlined version focused on coding tasks, providing faster and more cost-effective performance. Access to these models is being rolled out gradually, with ChatGPT Plus and ChatGPT Team subscribers getting immediate access, followed by ChatGPT Enterprise and ChatGPT Edu users next week. Free users will soon have access to OpenAI o1-mini as well.
- AI Chatbot
- Freemium • Proprietary
Related news
External links
- Introducing OpenAI o1-previewOpenAI Blog • Official source
- OpenAI’s new “reasoning” AI models are here: o1-preview and o1-miniArs Technica • Official source
- OpenAI o1 is the company's latest model for solving more complex tasksNeowin
- OpenAI launches new series of AI models with 'reasoning' abilitiesReuters
Comments
They're calling it "o1" to start version number as GPT were basically beta versions, with too much hallucination and non-sense. This version is an improvement but giving the results in their paper, by not that much. It seems we're reaching what AI/ML can offer as broad generating tools. Because AI/ML are very powerful tools for specific tasks, but as long as they push one-fits-all model for the sake of marketing as many people have many different interests, it won't be very good except parroting some basic web search. It's like asking a plumber to give a quantum mechanics course, he may have interesting things to say, he's not an expert and can talk nonsense. There is a line between strictly checking actual facts and generating pure creative content that they've yet to clearly define.
As long as they keep implementing too many guardrails, what it can actually achieve will be stifled.
Even 4o Mini is a fantastic model that is so much better than the old 3.5 Turbo. I remember when GPT4 was the limited premium model that reverted to 3.5 when you used up your free daily allowance. Now 4o Basic is the freemium and you get booted to 4o Mini after some time. But 4o Mini is still way better. I can't believe soon this o1 Mini will be the free model. Progress in this area is insane.
What a time to be alive, oh my god. GPT 4o only came out in May. I assume every few months for as long as we live, we will probably continue to see mind-blowing things.
Yes, but will they be GOOD things? I agree that AI blows something, but it ain't my mind.....
13% improvement in Mathematics Olympiad, and significant improvements in other areas as well. I'd say that's good any way you slice it.
Meh....