Lugel / Research

Research.

Notes on intelligence, computation, and verifiable systems.

All publications

2 pieces

Episode22 August 20262 min read

Move information. Check the work. Keep the record.

Research note22 August 20261 min read

Cuna applies inference optimizations to open models. A recorded GLM-5.2 test reached about 260 tok/s versus an OpenRouter route at 58–60 tok/s.