The biggest difference between o1 and the company's previous models is its chain-of-thought reasoning. Although not yet fully released, the preview and mini models have already far outperformed the gpt-4o in math, science, and coding tests.
The new model is the first of its kind, capable of reasoning in real time (just like a human).
What does your reasoning ability mean to users? "it's philippines mobile phone number really good, materially better," said one ceo with advanced access.
Create ai agents
create your own artificial intelligence agent in llm for free
start now
no credit card needed
when does gpt-5 go on sale?
Openaila latest version of llm was made public on september 12, 2024. The version included the o1-preview and o1-mini models.
Until its launch, predictions were very varied, and both users and journalists estimated that it could arrive in the summer of 2024 or 2026.
How smart is gpt-5?
Openai has presented a list of stem tests that demonstrate o1's reasoning ability:
performance similar to that of doctoral students on benchmark tests in physics, chemistry and biology.
Rank among the top 500 students in the qualifying test for the united states mathematical olympiad.
Ranking in the 89th percentile on codeforces, a competitive coding test.
You can read more about o1's reasoning ability in openai's research release.
Reasoning with openai o1
strawberry project
openai o1 was previously codenamed strawberry, with a strong component of mysticism and intrigue. "how strawberry works is a very well-kept secret, even in openai," an anonymous source explains to reuters.
The smaller version of this new ai was released on september 12, 2024 as part of a chatgpt update. Openai will likely use the larger version to generate training data for its llms, which could replace the need for large amounts of real-world data.
At an internal openai meeting on july 9, a demonstration was given of what could be the strawberry project, which was said to display human-like reasoning ability.
gpt-4 and gpt-5?
Openai ceo sam altman believes the world has only scratched the surface of ai. At the january 2024 world government summit, altman compared current openai models to the early days of mobile phones:
"the current technology we have is like that first mobile phone with the black and white screen that can only show those numbers... It just didn't do much... And at the time, [making a phone call] seems cool. And then it took us... Many decades from that to get to the iphones we have today... We're on this barely useful mobile phone."
although it will take time to go from the flip phone version of gpt to the iphone version, the o1 model brings us one step closer.
Openai ceo sam altman presents the launch of gpt-4 turbo from a dark stage.
1) greater reasoning ability
at the core of his general intelligence is o1's new reasoning ability. "perhaps the most important areas of progress revolve around reasoning abilities," altman tells gates. "right now, gpt-4 can only reason in extremely limited ways."