A new way to benchmark LLMs. Feel free not to trust me and go through the source code, I'm not stealing any API keys.