OpenAI's internal engineering benchmark shows model pass rates have stalled below 10% with GPT-5.5 scoring 1.7% · Digg