OpenAI's strongest reasoning model O3 released: AGI capacity skyrocketed close to human level

2024-12-31 23:16:31|Myriagame |source：minecraft skins

Today (December 21), OPENAI "Published on the 12th consecutive Christmas" finally ushered in an exciting finale. OpenAI launched a new product with a heavy uncle. It has so far upgraded the strongest frontier reasoning model -O3.

OpenAI claims that O3 is close to General Artificial Intelligence (AGI) under some conditions, and has two versions of O3 and streamlined O3-mini.

The assessment shows that O3 has a accuracy rate of nearly 47%higher than O1 in software engineering testing, and the accuracy of competition mathematics evaluation is 15%higher than that of O1. The accuracy rate in the test of doctoral experts in humans is nearly 13%higher than that of O1.

In AGI -related tests, O3's best score was 87.5, exceeding the level of human level 85, showing that the intelligence -like intelligence made breakthroughs.

In contrast, the evaluation results of the GPT-3 are 0%, the GPT-4O is 5%, and O3 has raised the results to 87.5%in one fell swoop.

OpenAI plans to release the O3 series model early next year. CEO Altman implies that it may not be launched to the public for a period of time. It is said that before it is officially released, there is a federal government test framework to guide monitoring and reduce the risks of such models.Essence

Beginning this Friday, OpenAI allows security researchers to register to visit O3 and O3-Mini previews.

• 

  National Service DNF Dark Ni

• 

  The new screenshot of the be

• 

  The 15th anniversary wallpap

• 

  3DM Xuanyuan Sinicization Gr

• 

  French magazine is the first

• 

  The sneak game "Republic" wi