GLM — Field Evidence
paperUnverified
GLM · GLM · adoption
Field note
Cross-Tokenizer On-Policy Distillation via Byte-Prefix Marginalization — Open-weight language models from different families exhibit complementary capabilities, motivating their consolidation into a compact student through on-policy distillation (OPD). However, full-vocabulary OP
collected 2026-07-27original 2026-07-24
Does this shift the US–China race?
Be the first to call it
Impact on the index
Open weights · minor
China +23Open +15
Directional contribution — before recency decay and per-type diminishing returns. How it works →
Related Front
US vs ChinaLikely
U.S. FrontiervsChina Open-Weight
Likely — capability gap narrowing on common tasks
Letters from the Front
Be the first to file a report.