OpenAI o3 is the next-generation reasoning model introduced by OpenAI in late December 2024, taking the cognitive capabilities of the “o” series to historic heights in science and software engineering.
Evolution compared to the o1 generation
o3 outperforms o1 by 20% in complex coding problems and stands out for a denser and more structured chain of thought, minimizing hallucinations to practically zero levels in formal proofs of theorems and computer security analysis.
Software Engineering
+20%
Direct improvement compared to o1 in refactoring and bug resolution tests.
Science and Logic
Border
Assessments of abstract reasoning with solving unseen problems.
Flexible Inference
Multilevel
Architecture designed to modulate the computing time according to the difficulty required.
For its optimized public access version, read our article on o3-mini and expand the vision in our OpenAI guide.
Cet article est également disponible en :