PostTrainBench - GLM modified verification protocol
Code
GLM release evaluation with three-run weighted average, zero-shot baseline fallback for failed scoring and LLM inspection replacing pattern-based external-API checks. Hardware and exact version are unreported; diagnostic rather than an assumed public v1.2/H100 result.
Top models (higher is better)
| Model | Score |
|---|---|
| GLM-5.3 | 39.8 |