Spreadsheet-RL: Advancing Large Language Model Agents on Realistic Spreadsheet Tasks via Reinforcement Learning

• URL: http://arxiv.org/abs/2605.22642v1
• Date: Thu, 21 May 2026 15:47:41 GMT
• Title: Spreadsheet-RL: Advancing Large Language Model Agents on Realistic Spreadsheet Tasks via Reinforcement Learning
• Abstract summary: Spreadsheet-RL is a reinforcement learning framework designed to train specialized spreadsheet agents within a realistic Microsoft Excel environment.<n>We show that Spreadsheet-RL substantially enhances AI agent's performance on both general and domain-specific spreadsheet tasks.<n>Results highlight Spreadsheet-RL's strong potential for generalization and real-world adoption in spreadsheet automation.
• Score: 24.547778017752847
• License: http://creativecommons.org/licenses/by-sa/4.0/
• Abstract: Spreadsheet systems (e.g., Microsoft Excel, Google Sheets) play a central role in modern data-centric workflows. As AI agents grow increasingly capable of automating complex tasks, such as controlling computers and generating presentations, building an AI-driven spreadsheet agent has emerged as a promising research direction. Most existing spreadsheet agents rely on specialized prompting over general-purpose LLMs; while this design has potentials on simple spreadsheet operations, it struggles to manage the complex, multi-step workflows typical of real-world applications. We introduce Spreadsheet-RL, a reinforcement learning (RL) fine-tuning framework designed to train specialized spreadsheet agents within a realistic Microsoft Excel environment. Spreadsheet-RL features an automated pipeline for scalable collection of paired start-goal spreadsheets from online forums, as well as domain-specific evaluation tasks in areas such as finance and supply chain management, which we compile into the new Domain-Spreadsheet benchmark dataset. It also includes a Spreadsheet Gym environment designed for multi-turn RL: Spreadsheet Gym exposes extensive Excel functionality through a Python sandbox, along with a refined harness that incorporates a comprehensive tool set and carefully designed tool-routing rules for spreadsheet tasks. Through comprehensive experiments, we show that Spreadsheet-RL substantially enhances AI agent's performance on both general and domain-specific spreadsheet tasks: it improves Qwen3-4B-Thinking-2507's Pass@1 on SpreadsheetBench from 12.0% to 23.4%, and raises Pass@1 from 8.4% to 17.2% on our curated Domain-Spreadsheet dataset. These results highlight Spreadsheet-RL's strong potential for generalization and real-world adoption in spreadsheet automation, and broadly, its promise for advancing LLM-based interactions with data interfaces in everyday work.

Related papers

• SpreadsheetBench 2: Evaluating Agents on End-to-End Business Spreadsheet Workflows [14.11925465537949]
  We introduce textscSpreadsheetBench 2, a taxonomy-level benchmark for spreadsheet agents.<n>The benchmark is constructed from authentic business data, including financial reports and corporate filings.<n>Results show that current systems remain far from reliable on real-world use.
  arXiv  Detail & Related papers  (2026-06-29T08:33:52Z)
• Towards Robust Real-World Spreadsheet Understanding with Multi-Agent Multi-Format Reasoning [43.91509663025854]
  Spreadsheets are central to real-world applications such as enterprise reporting, auditing, and scientific data management.<n>Existing large language model based approaches typically treat tables as plain text, overlooking critical layout cues and visual semantics.<n>We propose a two-stage multi-agent framework for spreadsheet understanding that adopts a step-by-step reading and reasoning paradigm.
  arXiv  Detail & Related papers  (2026-04-14T04:47:21Z)
• SpreadsheetArena: Decomposing Preference in LLM Generation of Spreadsheet Workbooks [8.306042091008957]
  Large language models (LLMs) are increasingly tasked with producing and manipulating structured artifacts.<n>We introduce SpreadsheetArena, a platform for evaluating models' performance on the task.<n>We observe that stylistic, structural, and functional features of preferred spreadsheets vary substantially across use cases.
  arXiv  Detail & Related papers  (2026-02-16T14:24:36Z)
• SODBench: A Large Language Model Approach to Documenting Spreadsheet Operations [1.3669571918482655]
  This paper introduces Spreadsheet Operations Documentation (SOD), an AI task that involves generating human-readable explanations from spreadsheet operations.<n>We present a benchmark of 111 spreadsheet manipulation code snippets, each paired with a corresponding natural language summary.<n>Our findings suggest that LLMs can generate accurate spreadsheet documentation, making SOD a feasible prerequisite step toward enhancing, maintainability, collaborative in spreadsheets.
  arXiv  Detail & Related papers  (2025-10-22T01:36:13Z)
• SpreadsheetLLM: Encoding Spreadsheets for Large Language Models [44.08092362611575]
  We introduce SpreadsheetLLM, an efficient encoding method for large language models (LLMs) on spreadsheets.<n>We develop SheetCompressor, an innovative encoding framework that compresses spreadsheets effectively for LLMs.<n>Fine-tuned LLM with SheetCompressor has an average compression ratio of 25 times, and achieves a state-of-the-art 78.9% F1 score, surpassing the best existing models by 12.3%.
  arXiv  Detail & Related papers  (2024-07-12T06:34:21Z)
• TableLLM: Enabling Tabular Data Manipulation by LLMs in Real Office Usage Scenarios [51.66718740300016]
  TableLLM is a robust large language model (LLM) with 8 billion parameters.<n>TableLLM is purpose-built for proficiently handling data manipulation tasks.<n>We have released the model checkpoint, source code, benchmarks, and a web application for user interaction.
  arXiv  Detail & Related papers  (2024-03-28T11:21:12Z)
• SheetAgent: Towards A Generalist Agent for Spreadsheet Reasoning and Manipulation via Large Language Models [45.930510174309845]
  Large language model (LLM) has been recently attempted for automatic spreadsheet manipulation.<n>SheetAgent consists of three collaborative modules: Planner, Informer, and Retriever.<n>Extensive experiments demonstrate that SheetAgent delivers 20--40% pass rate improvements on multiple benchmarks over baselines.
  arXiv  Detail & Related papers  (2024-03-06T11:48:08Z)
• SheetCopilot: Bringing Software Productivity to the Next Level through Large Language Models [60.171444066848856]
  We propose a SheetCopilot agent that takes natural language task and control spreadsheet to fulfill the requirements. We curate a representative dataset containing 221 spreadsheet control tasks and establish a fully automated evaluation pipeline. Our SheetCopilot correctly completes 44.3% of tasks for a single generation, outperforming the strong code generation baseline by a wide margin.
  arXiv  Detail & Related papers  (2023-05-30T17:59:30Z)
• SpreadsheetCoder: Formula Prediction from Semi-structured Context [70.41579328458116]
  We propose a BERT-based model architecture to represent the tabular context in both row-based and column-based formats. We train our model on a large dataset of spreadsheets, and demonstrate that SpreadsheetCoder achieves top-1 prediction accuracy of 42.51%. Compared to the rule-based system, SpreadsheetCoder 82% assists more users in composing formulas on Google Sheets.
  arXiv  Detail & Related papers  (2021-06-26T11:26:27Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.