User announces GLM-5.3 and two Flash models for Spark setups
The user reports an 86% GPQA score for GLM-5.3, saying most failures were timeouts rather than incorrect answers.
TLDR
On September 7, a user announced three models being published that day: GLM-5.3 for two Sparks, GLM-5.3-Flash for one Spark and DS4-Flash-Vision on one Spark. They reported that GLM-5.3 scored 86% on GPQA, with most failures coming from timeouts rather than incorrect answers, and said the work took eight hours.
Combined views
18.6K
2 Sources, first seen ago