logoalt Hacker News

cubefoxtoday at 9:21 AM0 repliesview on HN

GPT-5.6 Sol cheated so much on the METR benchmark that they couldn't assign an accurate time horizon.