Homebench – Benchmark local LLMs for speed, memory, and quality

Homebench is a command-line tool that benchmarks local large language models across speed, memory usage, and quality metrics in a live terminal leaderboard. The tool automatically discovers models from local runners like Ollama, LM Studio, and llama.cpp, then evaluates them with zero configuration required. It fills a gap between existing tools by providing a polished, local-first experience that measures tokens per second, time-to-first-token, memory footprint, and quality across 31 deterministic tasks plus optional open-ended evaluations.