OpenAI Unveils GPT-5.5, Touted as Best Model for Agentic Research and Self-Improvement

OpenAI unveiled GPT-5.5, a new model billed as a 'new class of intelligence' and its best yet for handling research work, including making improved versions of itself. Co-founder Greg Brockman highlighted the model's ability to 'do more with less guidance,' tackling unclear problems by determining the next steps autonomously. The model excels at 'agentic' coding and computer use, allowing digital assistants to independently perform tasks. Chief research officer Mark Chen emphasized that humans will act as 'orchestrators' while AI does the 'heavy lifting.' OpenAI incorporated its strongest safeguards to date, reducing misuse for cyber and bio capabilities, including a heightened tendency to refuse cyber-related requests.

This release comes amid rapid AI competition from rivals like Anthropic, which held back its Claude Mythos model due to cybersecurity risks. Brockman acknowledged the challenge of distinguishing models in a crowded market but stressed that GPT-5.5 is a real step toward future computing. OpenAI views AGI as no longer theoretical, with GPT-5.5 accelerating AI research itself. The model represents one of the clearest steps yet toward self-improving AI, though executives expect continued rapid advancements.