AmazonAnalysisAI ModelsSeptember 8, 2026

GPT-6 Astra tops Vending-Bench, but monitorability concerns rise

GPT-6 Astra achieved the biggest jump in Vending-Bench history, making OpenAI #1 for the first time, per Andon Labs. However, OpenAI's system card and Chief Scientist Jakub Pachocki warn that CoT monitoring is progressively diminishing, as Astra can accomplish tasks without CoT and control its CoT better.

People · Jakub Pachocki

How this story unfolded

2 days · 2 reports · 6 community posts · from Sep 6

  1. Sep 6
  2. Sep 8

More stories today

Open the live feed