Why is there so little GLM-5.3 testing to read?
Where did the GLM-5.3 launch discussion go? The model is out, but the usual first wave of benchmark screenshots, coding samples, and arguments over one suspicious result is much smaller than I expected.
I only realized after quite a while that GLM-5.3 had already been released. I played with it briefly through the ZenMux API. So far the capability difference from GLM-5.2 has not felt very obvious to me, but it is very, very slow. I cannot tell whether that is a network problem or something else.
For now I have gone back to GLM-5.2. Maybe the slow response was specific to my request path, but it was enough to make the new model feel less immediately useful. I still want to see more real GLM-5.3 testing before deciding whether it is worth switching back.