LLM-speed-benchmark results (GPT-4o mini slower than GPT-4o?)github.com/coder5432 pointscoder5432 years ago