OpenAI's New Reasoning Model Crosses a Threshold Researchers Didn't Expect
The latest frontier model solves multi-step problems that stumped its predecessors — and it's reshaping how labs think about scaling.
Ada Okonkwo
Senior AI Correspondent
A quiet leap
For the past two years, progress in large language models has felt incremental — a few points on a benchmark here, a faster inference path there. This week's release breaks that pattern. On a battery of multi-step reasoning tasks, the new model doesn't just edge past its predecessor; it clears problems that an entire generation of systems failed outright.
Researchers describe the change as qualitative rather than quantitative. "It's not that it's 5% better," one engineer told AfroEuropa. "It's that a class of problems went from impossible to routine."
Keep reading
One newsletter, two continents
The Bridge brings you the tech, startups, and leaders moving between Africa and Europe — one sharp email each morning. No spam, unsubscribe anytime.
