OpenAI said the biggest leap is in agentic coding and computer. On Terminal-Bench 2.0, which tests complex command-line ...