GLM — Field Evidence
redditUnverified
GLM · GLM · incident
Field note
We released an abliterated + fine-tuned GLM-5.2. High scores on adversarial benchmarks while keeping coding performance. — We just shipped abliterated-model-large. It is GLM-5.2 with the refusal directions removed, then fine-tuned specifically for long adversarial and agent-style
collected 2026-07-26original 2026-07-25
Does this shift the US–China race?
Be the first to call it
Impact on the index
Regulation · major
China -110US +30
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.