OpenAI has developed an internal model under code name "Strawberry" (formerly "Q") that scores over 90% on MATH, a championship-level mathematics benchmark, and has demonstrated capability on complex science and math problems beyond current commercial models. The project targets agentic use cases including autonomous web research with computer-using agents and evaluation against software and machine learning engineering tasks. The development suggests OpenAI is advancing multi-step reasoning and autonomy capabilities, with specific focus on practical agentic workflows. For financial institutions, this represents evidence of progress in reasoning-time compute and agentic reliability, two structural questions that bear directly on the productivity and deployment economics of frontier models.
206 other tracked developments touching the same claims.