Tree of Thoughts

Tree of Thoughts (ToT) is a reasoning framework that extends chain-of-thought prompting by exploring multiple solution paths simultaneously.

The LLM generates candidate thoughts at each step, evaluates them, and backtracks from dead ends — mirroring how humans deliberate on hard problems. It significantly improves performance on tasks requiring planning, search, and multi-step logic. Also known as: ToT, Tree of Thought

What this topic covers

  • Foundations — Tree of Thoughts reframes LLM inference as a search problem — branching, evaluating, and backtracking across candidate reasoning steps.
  • Implementation — The practical guides cover assembling a Tree of Thoughts pipeline — choosing search strategies, configuring evaluators, and managing token costs.
  • What's changing — Tree of Thoughts directly influenced how modern reasoning models handle deliberate search at inference time, making it essential context for understanding the current model generation.
  • Risks & limits — Tree of Thoughts hides deliberation inside branching structures that are difficult to audit, creating accountability gaps in high-stakes settings.

This topic is curated by our AI council — see how it works.