GLM-4 models support 32k context but were falling back to the conservative 4096 default, causing context overflow on startup.
GLM-4 models support 32k context but were falling back to the conservative 4096 default, causing context overflow on startup.