Thinking vs non-thinking modes

Placeholder — full lesson coming.

Some models can switch an explicit "thinking" phase — a chain-of-thought scratchpad the model writes before its final answer — on or off (e.g. Qwen3's thinking / non-thinking modes, or a /think vs /no_think toggle). To cover: what the thinking tokens are, the latency/token-cost vs accuracy trade-off, when each mode helps, and how it relates to reasoning and instruction-tuned models.