Greenblatt: 3x Depth Gain Comes From Architecture, Not Parameter Scaling

Former Anthropic researcher Ryan Greenblatt argues that a 3x gain in reasoning depth of a new OpenAI model stems from architectural changes such as looped transformers rather than parameter scaling, which he estimates would require roughly 81x more parameters. Critics, including xuanalogue, challenged the math behind his conversion.

2026-09-04 ~ 2026-09-04 · 4 related posts

• Greenblatt: 3x jump in model depth points to architecture changes, not scaling — RyanGreenblatt · 2026-09-04
• Is recurrent depth the real driver? Researchers debate 3x depth vs 81x params math — xuanalogue · 2026-09-04
• 3x depth equals 3x params? Skeptic challenges the 81x equivalence math — xuanalogue · 2026-09-04
• Scaling depth 3x means 81x params: Ryan Greenblatt on looped transformer scaling — RyanGreenblatt · 2026-09-04