loading
loading
Instead of one reasoning chain, the model explores several branching paths and keeps the most promising ones.
Tree of Thoughts (ToT) generalizes chain-of-thought: at each step the model proposes multiple candidate next thoughts, scores them, and searches the tree (breadth- or depth-first), backtracking from dead ends. It buys accuracy on problems where a single greedy chain commits early to a wrong turn — at the cost of many more tokens and latency. Example: solving a logic puzzle or game where the model evaluates three possible moves per step and prunes the losers.
Plainly
Think of Tree of Thoughts as a simple recipe for doing the work better. Instead of one reasoning chain, the model explores several branching paths and keeps the most promising ones.
In practice
Use it when you need a repeatable method instead of guessing from vibes. In practice, define the owner, input, output, and failure mode before you rely on it.