---
title: "Language Agent Tree Search Unifies Reasoning Acting and Planning in 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 2026-11-12; icon by-4.0.png, http://creativecommons.org/licenses/by/4.0/); note is our own summary'
source: https://iinventedlight.com/library/notes/2310-04406-language-agent-tree-search.md
source_license: 'CC BY 4.0'
attribution: 'Summary by iinventedlight (AI-assisted). Source: Andy Zhou, Kai Yan, Michal Shlapentokh-Rothman, Haohan Wang, Yu-Xiong Wang, Language Agent Tree Search Unifies Reasoning Acting and Planning in Language Models (CC BY 4.0).'
author: 'iinventedlight (AI-assisted)'
date: 2026-11-12
status: published
approved_by: "Cole Withington, 2026-11-12 14:44 PT, Librarian chat widget t51s316: \"Yes, publish all 100 second-list arXiv notes\""
checked_on: 2026-11-12
as_of: 2026-11-12
confidence: medium
untested: "Not reproduced. Abs page and HTML key sections read 2026-11-12. Named code URLs checked for HTTP 200; code not run."
category: arxiv
topic_area: 'ai-research: agents'
tags: arxiv, agents, tree-search, planning, lats
arxiv_id: 2310.04406
alias: 2310-04406-language-agent-tree-search
sources:
  - id: S1
    title: "Language Agent Tree Search Unifies Reasoning Acting and Planning in Language Models"
    publisher: arXiv
    authors: "Andy Zhou, Kai Yan, Michal Shlapentokh-Rothman, Haohan Wang, Yu-Xiong Wang"
    url: https://arxiv.org/abs/2310.04406
    version_read: "v3 (6 Jun 2024)"
    accessed: 2026-11-12
    license: 'CC BY 4.0'
---
# Language Agent Tree Search Unifies Reasoning Acting and Planning in Language Models

## Summary
Language Agent Tree Search (LATS) is a framework that combines language-model reasoning, acting, and planning. It folds Monte Carlo Tree Search into an agent loop, with language-model value functions and self-reflection, and it uses environment feedback during search. The authors evaluate on programming, interactive question answering, web navigation, and math. They report 92.7% pass@1 on HumanEval with GPT-4, and an average score of 75.9 on WebShop with GPT-3.5 that they compare with gradient-based fine-tuning. Code can be found at https://github.com/lapisrocks/LanguageAgentTreeSearch. [S1]

## Claims
- id: C1
  claim: "LATS combines reasoning, acting, and planning by using Monte Carlo Tree Search in a language-agent loop."
  source: S1
  evidence: "integrate Monte Carlo Tree Search into LATS to enable LMs as agents"
- id: C2
  claim: "With GPT-4, LATS reaches 92.7% pass@1 on HumanEval."
  source: S1
  evidence: "pass@1 accuracy (92.7%) for programming on HumanEval with GPT-4"
- id: C3
  claim: "Code is released at a public GitHub repository."
  source: S1
  evidence: "https://github.com/lapisrocks/LanguageAgentTreeSearch"

## Verification
- Code available: Paper: https://github.com/lapisrocks/LanguageAgentTreeSearch, HTTP 200 on 2026-11-12, redirecting to https://github.com/andyz245/LanguageAgentTreeSearch. Not run.
- Reproduced: not reproduced (2026-11-12). Code not run.
- Related work: Search-style prompting is in [2305-10601-tree-of-thoughts](https://iinventedlight.com/library/notes/2305-10601-tree-of-thoughts/). Acting with tools and observations is in [2210-03629-react-reasoning-and-acting](https://iinventedlight.com/library/notes/2210-03629-react-reasoning-and-acting/). HumanEval: [2107-03374-humaneval](https://iinventedlight.com/library/notes/2107-03374-humaneval/). WebShop is in this batch.
- Limits: Headline numbers use GPT-4 or GPT-3.5 in specific domains. Tree search is more expensive than single-pass ReAct-style acting.
- License: CC BY 4.0 on the abs page (2026-11-12; icon by-4.0.png; http://creativecommons.org/licenses/by/4.0/). Submitted 6 Oct 2023 (v1); last revised v3 (6 Jun 2024). Sources: [S1] https://arxiv.org/abs/2310.04406, v3 (6 Jun 2024). checked_on: 2026-11-12.

## When to apply
Use when asking how tree search can be combined with ReAct-style acting, or how LATS was tested on HumanEval and WebShop.

## Facts
- id: F1
  claim: "LATS uses Monte Carlo Tree Search, language-model value functions, and environment feedback in one agent framework."
  confidence: medium
  date: 2026-11-12
  sources: [S1]
  license: CC BY 4.0
- id: F2
  claim: "The authors report 92.7% pass@1 on HumanEval with GPT-4 and an average score of 75.9 on WebShop with GPT-3.5."
  confidence: medium
  date: 2026-11-12
  sources: [S1]
  license: CC BY 4.0
- id: F3
  claim: "Code was pointed to https://github.com/lapisrocks/LanguageAgentTreeSearch."
  confidence: medium
  date: 2026-11-12
  sources: [S1]
  license: CC BY 4.0

## Sources
- [S1] arXiv 2310.04406, https://arxiv.org/abs/2310.04406. License: CC BY 4.0. v3 (6 Jun 2024). Accessed 2026-11-12.

License: this note is CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Summary by iinventedlight (AI-assisted). Paper is CC BY 4.0.

---

Agents: [/agents.md](/agents.md) | [/.well-known/agent-library.json](/.well-known/agent-library.json).
