Tree of Thought (ToT) is a prompting strategy where an AI doesn't just generate one answer — it explores multiple reasoning branches, evaluating different paths before committing to a conclusion. Think of it like a chess player who mentally simulates several moves ahead rather than just reacting.

Standard prompting asks a model to go from question to answer in one pass. ToT lets the model backtrack, compare, and prune bad paths, which improves performance on complex multi-step problems like math proofs or logical puzzles.

The catch is cost. Each branch requires additional model calls, so ToT can multiply compute usage quickly. Recent research shows these reasoning budgets are often inelastic — throwing more compute at a problem doesn't always improve results proportionally, meaning smarter routing of reasoning effort matters as much as the total budget itself.