This workshop builds "model literacy” for anyone who has browsed OpenRouter, Hugging Face, or vendor pages and wondered what the specs actually imply in practice. Participants will learn how to interpret the most common model labels and tradeoffs—base vs instruction-tuned (and what “chat” really means), quantized vs full precision, distilled vs original, and mixture-of-experts vs dense models. We’ll connect these concepts to real decisions: capability vs cost, latency vs quality, context length vs reliability, and parameter count vs “felt” performance. We’ll also demystify evaluation: what benchmark scores do and don’t tell you, how to spot mismatched evals (chat vs base, tool-use vs no tool-use), and how to run quick, task-relevant checks using input→output behavior. The goal is that you leave able to compare models confidently and pick the right one for a given use case.