GLM feels faster and more reliable in my experience. Anthropic and OpenAI models would hem and haw or straight up timeout during peak times.