How to Call 4 Free LLM Models from One OpenAI-Compatible Endpoint (Tested October 2026)