Every serious benchmarking effort eventually outgrows its GitHub repo name. The open LLM leaderboard, MLPerf inference, the llama.cpp speedrun threads, the tokens-per-second tables that circulate on every developer forum whenever a new model drops — these are the resources practitioners actually use. They are not maintained by the companies being benchmarked. They exist because someone decided the number mattered and built the infrastructure to track it honestly. bitspeed.org is available for that project.