Thinking Machines: more loop passes boost no-CoT capability without hurting monitorability

thinkymachines · x · 2026-11-21

Thinking Machines reported findings on looped reasoning's monitorability: no-CoT capability rises with more passes; at this scale, the chain of thought gets shorter but stays legible, with no meaningful drop in monitorability. The team is also releasing a variable-loop research model so others can run their own experiments.

Original post →

More from Models

Models channel →