Maybe you have slow LLM model, maybe your network latency is very big, because al these models are in US