📈 AI agents can now handle tasks twice as long every 3.6 months, says METR

AI evaluator METR just released its biggest capability update yet, testing 15 models up to Opus 5.5 and GPT-6 Astra. The length of tasks top AI agents can finish is doubling roughly every 3.6 months. In March 2025 the best model managed about 1-hour tasks; today's frontier models hit up to 27 hours on METR's tougher verifiable tasks. The weak spot: when there's no clear feedback, AI agents still show worse judgment than human experts.

⭐ The work AI can do on its own is growing fast, but "is this actually right?" still needs a human.

#AI