Posts

I Benchmarked 27 AI Models on My MacBook Pro M5 Max — Speed, Quality, and Agentic

27 open-source LLMs benchmarked on M5 Max 128GB: speed (tok/s), quality (ARC, GSM8K, IFEval), and agentic capability (14 terminal-bench tasks). Most models can't even produce valid JSON.