Distillation trains open-source rivals, but hidden chains of thought may widen the gap

Closed-source labs are effectively training their own open-source replacements via distillation, threatening their moats. However, as frontier models stop exposing full chains of thought in plain text, distillation is becoming less effective, and the open-closed gap may keep widening.

2026-09-29 ~ 2026-09-30 · 2 related posts

• Hidden CoT and failing distillation: open-weight models set to fall further behind frontier labs — toptickcrypto · 2026-09-29
• Every closed model is training its open replacement: does distillation kill the moat? — alex_verem · 2026-09-30