New OpenAI o3 model: AI at its best | Chat OpenAI | ChatGPT OpenAI | OpenAI API | Turtles AI

OpenAI unveiled the o3 model, an innovation in AI that raises the standards for reasoning, coding and academic problem solving. With unprecedented results in areas such as mathematics and software engineering, o3 redefines AI capabilities.

Key points:

• Unprecedented mathematical performance: o3 achieves 96.7 percent accuracy on the AIME 2024 benchmark, outperforming predecessors.
• Progress in AI: the model solves an unsolved ARC task for the first time.
• Optimized efficiency: more power with lower consumption of computational resources.
• Advanced applications: o3 proves versatile in creative, scientific and engineering fields.

OpenAI has unveiled its new AI model, o3, which represents a qualitative leap from previous models in the o series. With advanced capabilities in reasoning, complex problem solving and content generation, o3 redefines standards in areas such as mathematics, science and software engineering. The model’s accuracy is reflected in its results on key benchmarks: on the AIME 2024, a high-level math test, o3 achieved an accuracy of 96.7 percent, outperforming both o1 (83.3 percent) and the preliminary version of o1 (56.7 percent). These numbers testify to the model’s ability to deal with structured and complex problems, solidifying its position as one of the most advanced tools for symbolic computation.

In the scientific domain, the model demonstrated outstanding comprehension, achieving 87.7 percent accuracy on doctoral-level questions from the GPQA Diamond set. Even more impressive is its performance on frontier questions, such as the EpochAI Frontier Math benchmark, where it set a new record with an accuracy of 25.2%, up from the previous 2.0%. These results underscore the model’s ability to tackle specialized and abstract problems, pushing the limits of AI to new frontiers.

Another significant milestone was achieved in the ARC benchmark, which measures abstract reasoning and generalization ability. Here, o3 is the first model to solve one of the remaining unsolved problems, demonstrating skills never before achieved in domains related to general AI. This achievement marks a milestone in the development of AI capable of tackling complex logical challenges and adapting to scenarios never before encountered.

An additional strength of the o3 model is its efficiency. Despite its enhanced capabilities, it requires fewer computational resources than previous versions. Its optimized structure enables it to adapt to concrete application contexts, easily integrating structured outputs, advanced functions, and responses to intricate scenarios. This feature makes it ideal for a wide range of applications, from supporting scientific research to creative content creation and complex systems development.

The unveiling of o3, scheduled for a phased release starting in January, marks a major step in the AI landscape. The model embodies a new gold standard for the future of the o series, redefining expectations for the capabilities of AI systems.