Bill The Investor|Apr 11, 2026 04:49
GPT-5.4-xhigh is completely done for
In METR's time span test, it only lasted 5.7 hours
Opus 4.6 is close to 12 hours
According to METR, GPT-5.4 'produces reward hacks at an unusually frequent rate'
Share To
HotFlash
APP
X
Telegram
CopyLink