Databricks Engineer Claims GLM-5.3 Tops Speed and Coding Tests

Yuchen Jin, Research Engineer at Databricks, posted that GLM-5.3 reaches 310 tok/s on the company's inference platform. He stated this leads both speed and latency. Jin also wrote that GLM-5.3 ranks as the strongest open-source model on Databricks internal coding benchmark and competes with Fable 5 and Opus 4.8. The post includes a screenshot of a Z AI benchmarking dashboard titled GLM-5.3 (max).