Zhidongxi reported on September 13 that OpenAI’s new model was launched without warning: the o1 series can conduct general complex reasoning, and each answer takes longer to think.
, When solving physics problems at the doctoral level, GPT-4o still “failed” with 59.
5 points, and o1 jumped to the “excellent stage” and directly achieved 92.
8 points! That’s right, the legendary strawberry has finally come to meet you!, 
,” CEO Ultraman calls it the beginning of a new paradigm: artificial intelligence that can perform general complex reasoning.
Starting today, o1-preview will be available to all Plus and Team users in ChatGPT and to tier 5 developers in the API.
, At the same time, OpenAI also released o1-mini-a cost-effective reasoning model that is very good at STEM, especially mathematics and coding.
, specifically, the o1 series is OpenAI’s first model to undergo intensive learning training.
Before outputting answers, a long thought chain will be generated to enhance the model’s capabilities.
In other words, the longer the internal thought chain and the longer o1 thinks, the better the model will perform on reasoning tasks.
How strong is O1? CEO Ultraman gave the answer straight: In the just-concluded 2024 IOI Informatics Olympics questions, the fine-tuned version of o1 scored 213 points with 50 attempts per question, ranking among the top 49% of human players.
If it is allowed to try 10000 times per question, it will score 362.
14 points, which is higher than the gold medal player threshold and can win the gold medal.
In addition, it ranks among the top 89% of Competitive Programming Questions (Codeforces) and ranks among the top 500 students in the United States in the American Mathematical Olympics (AIME) qualifying questions.
, Return to the first electric network home page>,.
