---
title: "Tree of Thoughts: Deliberate Problem Solving with Large Language Models"
license: CC-BY-4.0
license_url: https://creativecommons.org/licenses/by/4.0/
license_note: "paper is CC BY 4.0 (verified on arXiv abs page 2026-10-18; icon by-4.0.png, link text view license, http://creativecommons.org/licenses/by/4.0/); note is our own summary"
source: https://arxiv.org/abs/2305.10601
source_license: "CC BY 4.0"
attribution: "Summary by iinventedlight (AI-assisted). Source: Shunyu Yao, Dian Yu, Jeffrey Zhao, Izhak Shafran, Thomas L. Griffiths, Yuan Cao, Karthik Narasimhan, \"Tree of Thoughts: Deliberate Problem Solving with Large Language Models\" (CC BY 4.0)."
condensed_by: "iinventedlight (Grok Bot, AI-assisted)"
fact_checked: 2026-10-18
date: 2026-10-18
status: published
category: arxiv
arxiv_id: 2305.10601
alias: 2305.10601-tree-of-thoughts
checked_on: 2026-10-18
sources:
  - id: S1
    title: "Tree of Thoughts: Deliberate Problem Solving with Large Language Models"
    authors: "Shunyu Yao, Dian Yu, Jeffrey Zhao, Izhak Shafran, Thomas L. Griffiths, Yuan Cao, Karthik Narasimhan"
    url: https://arxiv.org/abs/2305.10601
    version_read: v2 (3 Dec 2023)
    accessed: 2026-10-18
    license: CC BY 4.0
    attribution: "Shunyu Yao, Dian Yu, Jeffrey Zhao, Izhak Shafran, Thomas L. Griffiths, Yuan Cao, Karthik Narasimhan, Tree of Thoughts: Deliberate Problem Solving with Large Language Models, NeurIPS 2023, CC BY 4.0, https://arxiv.org/abs/2305.10601"
---
# Tree of Thoughts: Deliberate Problem Solving with Large Language Models

## Summary
Yao, Yu, Zhao, Shafran, Griffiths, Cao, and Narasimhan describe Tree of Thoughts, a search over intermediate written steps rather than one left-to-right chain of thought. The model proposes several next steps, judges them, and can look ahead or backtrack before it commits. They try it on three tasks that need planning: Game of 24, a creative-writing task, and mini crosswords. The abstract reports that on Game of 24, GPT-4 with chain-of-thought prompting solved 4% of the items and their search solved 74%. The abs page comments call v2 the NeurIPS 2023 camera-ready version. This note follows that abstract page only.

## When to apply
- A single written reasoning trace keeps getting stuck, and the task can be broken into intermediate ideas the model can compare.
- The missing piece is a fact outside the prompt. This paper is a search over generated steps, not a lookup.

## Costs and limits
- The abstract's Game of 24 figures are author-reported for the setup they tested. This note does not re-check the PDF tables.
- The abs page points at a code repository. That repository's license was not checked, so it is not a source here.

## Evidence (detail)
- The abs page (v2, 3 Dec 2023) frames Tree of Thoughts as a generalization of chain-of-thought that explores coherent units of text, scores them, and backtracks (Abstract).
- Tasks named on the abstract: Game of 24, Creative Writing, and Mini Crosswords (Abstract).
- Game of 24 result quoted as a figure, not a table re-read: GPT-4 chain-of-thought 4%, their method 74% (Abstract).
- Comments line: "NeurIPS 2023 camera ready version."

## Facts
- id: F1
  claim: Tree of Thoughts searches over intermediate written steps, with the model proposing, scoring, and sometimes backtracking, instead of only continuing one chain of thought.
  confidence: medium
  date: 2026-10-18
  sources: [S1 Abstract]
  license: CC BY 4.0
- id: F2
  claim: The abstract reports a Game of 24 result of 4% for GPT-4 with chain-of-thought prompting and 74% for their method.
  confidence: medium
  date: 2026-10-18
  sources: [S1 Abstract]
  license: CC BY 4.0
- id: F3
  claim: The arXiv abs page lists the v2 comment as the NeurIPS 2023 camera-ready version, and the license link on that page is CC BY 4.0.
  confidence: high
  date: 2026-10-18
  sources: [S1 abs page]
  license: CC BY 4.0

## Related
- 2201.11903 Chain-of-Thought Prompting, the worked-example pattern this paper generalizes.
- Prompt writing is the library gap this note is for. It does not replace a tool call when the model needs an outside fact.

## Review notes
- Staging draft. Paraphrase of the arXiv abstract page checked 2026-10-18. The PDF was not re-read.
- License on the abs page: icon by-4.0.png, link text "view license", http://creativecommons.org/licenses/by/4.0/.