After a 12-day technology sharing live broadcast event, OpenAI released the next-generation inference model o3, as well as a streamlined version of o3-mini optimized for specific tasks. o3 has made breakthrough progress in multiple benchmark tests, its performance significantly surpasses the previous generation model o1, and even approaches the level of artificial general intelligence (AGI) in some aspects. This release has attracted widespread attention in the industry and is believed to have a profound impact on future programming methods and programmers' working models.
After 12 days of technology sharing live broadcasts, OpenAI released its next-generation inference model o3 on the last day, which is an upgraded version of the o1 inference model released earlier. The o3 model series includes two versions: o3 and o3-mini, of which o3-mini is a smaller, streamlined model fine-tuned for specific tasks. OpenAI stated that the o3 model can come close to achieving general artificial intelligence (AGI) under certain conditions, that is, artificial intelligence that can complete any task that humans can complete.
In the ARC-AGI graphical logical inference benchmark, the o3 model achieved record-breaking scores, scoring 75.7% in the low-compute scenario, while in the high-compute test it reached 87.5%, surpassing the benchmark that marks reaching human levels. Threshold 85%. In comparison, the o1 model scores only between 25% and 32%, and o3 performs almost three times better than o1. On the world-famous coding competition platform Codeforces, o3 achieved a score of 2727, while o1 scored only 1891.
Fu Sheng, chairman of Cheetah Mobile or Orion Star, said that the release of OpenAI o3 heralds the coming of an era when everyone is a programmer. Users do not need to be proficient in Python or C language to write programs. They only need to put forward requirements and the big prediction model can help. Complete the programming work. Fu Sheng believes that the release of o3 marks that the programming ability of large language models surpasses 99.9% of programmers. In the Codeforces world-class programming competition, o3 achieved the top result of 175th place, while o1 only defeated more than 90% of programmers. Programmers, GPT-4o only defeated 11% of programmers before.
OpenAI plans to officially release the o3 model at the end of January next year. Fu Sheng pointed out that although programmers will not disappear completely, their work will shift more to understanding user needs and building large logic, and the work of converting needs into code will be largely completed by AI. This release heralds the wider application of AI in the field of programming and may also change the way programmers work.
The release of the o3 model marks a significant progress in artificial intelligence technology, and its powerful reasoning and programming capabilities will have a profound impact on various fields. In the future, with the continuous development and improvement of technology, we can expect artificial intelligence to play a greater role in more fields and bring greater convenience to human society.