Spreadsheet-RL: Advancing Large Language Model Agents on Realistic Spreadsheet Tasks via Reinforcement Learning
Mirrored from Hugging Face Daily Papers for archival readability. Support the source by reading on the original site.
Spreadsheet-RL is an RL fine-tuning framework and benchmarking environment designed to improve LLM agent performance on complex, multi-step spreadsheet tasks within Microsoft Excel.</p>\n","updatedAt":"2026-05-22T02:01:02.085Z","author":{"_id":"6039478ab3ecf716b1a5fd4d","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/6039478ab3ecf716b1a5fd4d/_Thy4E7taiSYBLKxEKJbT.jpeg","fullname":"taesiri","name":"taesiri","type":"user","isPro":true,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":303,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.8628856539726257},"editors":["taesiri"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/6039478ab3ecf716b1a5fd4d/_Thy4E7taiSYBLKxEKJbT.jpeg"],"reactions":[{"reaction":"🔥","users":["Mingyuan1997","BiboyQG"],"count":2}],"isReport":false}},{"id":"6a0fba89a641d26c3b2f91e4","author":{"_id":"65d3b7ec8f6b98b34ee6bbe3","avatarUrl":"/avatars/53c2d4e4746147fc2559435d252e8951.svg","fullname":"Mingyuan Wu","name":"Mingyuan1997","type":"user","isPro":true,"isHf":false,"isHfAdmin":false,"isMod":false,"isUserFollowing":false},"createdAt":"2026-05-22T02:08:09.000Z","type":"comment","data":{"edited":false,"hidden":false,"latest":{"raw":"\n\n\n\n\n","html":"<p><a href=\"https://cdn-uploads.huggingface.co/production/uploads/65d3b7ec8f6b98b34ee6bbe3/Kd-QsKT8UZKXLsYn4kA-R.png\" rel=\"nofollow\"><img src=\"https://cdn-uploads.huggingface.co/production/uploads/65d3b7ec8f6b98b34ee6bbe3/Kd-QsKT8UZKXLsYn4kA-R.png\" alt=\"10bdb50b0e36775dd5ed9da6c2f0c53\"></a></p>\n<p><a href=\"https://cdn-uploads.huggingface.co/production/uploads/65d3b7ec8f6b98b34ee6bbe3/9h-d-8o0gOy-BzJ1xfrpX.png\" rel=\"nofollow\"><img src=\"https://cdn-uploads.huggingface.co/production/uploads/65d3b7ec8f6b98b34ee6bbe3/9h-d-8o0gOy-BzJ1xfrpX.png\" alt=\"image\"></a></p>\n","updatedAt":"2026-05-22T02:08:09.946Z","author":{"_id":"65d3b7ec8f6b98b34ee6bbe3","avatarUrl":"/avatars/53c2d4e4746147fc2559435d252e8951.svg","fullname":"Mingyuan Wu","name":"Mingyuan1997","type":"user","isPro":true,"isHf":false,"isHfAdmin":false,"isMod":false,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.3536166250705719},"editors":["Mingyuan1997"],"editorAvatarUrls":["/avatars/53c2d4e4746147fc2559435d252e8951.svg"],"reactions":[{"reaction":"🚀","users":["sta1nlu"],"count":1}],"isReport":false}},{"id":"6a0fd0b0452c81b5b9102eae","author":{"_id":"65d3b7ec8f6b98b34ee6bbe3","avatarUrl":"/avatars/53c2d4e4746147fc2559435d252e8951.svg","fullname":"Mingyuan Wu","name":"Mingyuan1997","type":"user","isPro":true,"isHf":false,"isHfAdmin":false,"isMod":false,"isUserFollowing":false},"createdAt":"2026-05-22T03:42:40.000Z","type":"comment","data":{"edited":false,"hidden":false,"latest":{"raw":"https://spreadsheet-rl.github.io/","html":"<p><a href=\"https://spreadsheet-rl.github.io/\" rel=\"nofollow\">https://spreadsheet-rl.github.io/</a></p>\n","updatedAt":"2026-05-22T03:42:40.356Z","author":{"_id":"65d3b7ec8f6b98b34ee6bbe3","avatarUrl":"/avatars/53c2d4e4746147fc2559435d252e8951.svg","fullname":"Mingyuan Wu","name":"Mingyuan1997","type":"user","isPro":true,"isHf":false,"isHfAdmin":false,"isMod":false,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.28882917761802673},"editors":["Mingyuan1997"],"editorAvatarUrls":["/avatars/53c2d4e4746147fc2559435d252e8951.svg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2605.22642","authors":[{"_id":"6a0fb81aa53a61ce2e422c07","name":"Banghao Chi","hidden":false},{"_id":"6a0fb81aa53a61ce2e422c08","name":"Yining Xie","hidden":false},{"_id":"6a0fb81aa53a61ce2e422c09","name":"Mingyuan Wu","hidden":false},{"_id":"6a0fb81aa53a61ce2e422c0a","name":"Jingcheng Yang","hidden":false},{"_id":"6a0fb81aa53a61ce2e422c0b","name":"Jize Jiang","hidden":false},{"_id":"6a0fb81aa53a61ce2e422c0c","name":"Zhaoheng Li","hidden":false},{"_id":"6a0fb81aa53a61ce2e422c0d","name":"Shengyi Qian","hidden":false},{"_id":"6a0fb81aa53a61ce2e422c0e","name":"Minjia Zhang","hidden":false},{"_id":"6a0fb81aa53a61ce2e422c0f","name":"Klara Nahrstedt","hidden":false},{"_id":"6a0fb81aa53a61ce2e422c10","name":"Rui Hou","hidden":false},{"_id":"6a0fb81aa53a61ce2e422c11","name":"Xiangjun Fan","hidden":false},{"_id":"6a0fb81aa53a61ce2e422c12","name":"Hanchao Yu","hidden":false}],"publishedAt":"2026-05-21T00:00:00.000Z","submittedOnDailyAt":"2026-05-22T00:00:00.000Z","title":"Spreadsheet-RL: Advancing Large Language Model Agents on Realistic Spreadsheet Tasks via Reinforcement Learning","submittedOnDailyBy":{"_id":"6039478ab3ecf716b1a5fd4d","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/6039478ab3ecf716b1a5fd4d/_Thy4E7taiSYBLKxEKJbT.jpeg","isPro":true,"fullname":"taesiri","user":"taesiri","type":"user","name":"taesiri"},"summary":"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.\n 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.","upvotes":22,"discussionId":"6a0fb81aa53a61ce2e422c13","githubRepo":"https://github.com/Spreadsheet-RL/Spreadsheet-RL","githubRepoAddedBy":"user","ai_summary":"Spreadsheet-RL is a reinforcement learning framework that trains specialized spreadsheet agents in realistic Excel environments, improving AI agent performance on both general and domain-specific spreadsheet tasks through automated data collection and domain-specific benchmarks.","ai_keywords":["reinforcement learning","fine-tuning","spreadsheet agents","Microsoft Excel","automated pipeline","domain-specific evaluation","Domain-Spreadsheet benchmark","Spreadsheet Gym","multi-turn RL","tool-routing rules","Pass@1"],"githubStars":3,"organization":{"_id":"65448bef5b5d9185ba3202b9","name":"UIUC-CS","fullname":"University of Illinois at Urbana-Champaign","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/65448b21fcb96b8b48733729/ycqcXFayMTTD_KpE37067.jpeg"}},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"65614a38b29be3f5b69f296f","avatarUrl":"/avatars/9f000f695714e7bf0252648dbe2cfec7.svg","isPro":false,"fullname":"Jize Jiang","user":"jizej","type":"user"},{"_id":"65d3b7ec8f6b98b34ee6bbe3","avatarUrl":"/avatars/53c2d4e4746147fc2559435d252e8951.svg","isPro":true,"fullname":"Mingyuan Wu","user":"Mingyuan1997","type":"user"},{"_id":"65de7628deee79773f0f46f6","avatarUrl":"/avatars/6c509dbe96e47b47271eb74178c1c9ba.svg","isPro":false,"fullname":"Kai Yan","user":"kaiyan289","type":"user"},{"_id":"680cfb7c0b50df51fe75424b","avatarUrl":"/avatars/febb2dfd80676a8ee093d8c4e738bc44.svg","isPro":false,"fullname":"Yining Xie","user":"eileenxyn","type":"user"},{"_id":"641c5662c3983aa949110c76","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/641c5662c3983aa949110c76/VhnS7jffwfyRMfuL-qU_e.jpeg","isPro":false,"fullname":"Banghao Chi","user":"BiboyQG","type":"user"},{"_id":"5ee24411636bdb3834e2da24","avatarUrl":"/avatars/0e12a0b19ada4138ddf808501ba68423.svg","isPro":false,"fullname":"Zhiqing Sun","user":"zhiqings","type":"user"},{"_id":"65a7154ada9f6df141454f20","avatarUrl":"/avatars/1161a72aa28e3915786267bc5f288c07.svg","isPro":false,"fullname":"Haozhen Zheng","user":"zoezheng126","type":"user"},{"_id":"68ea0b47f32cc8ded49e8530","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/JB2gzI4apd6zA5Ix1wWgS.png","isPro":false,"fullname":"Xiong Tianhu","user":"tianhux2","type":"user"},{"_id":"6606434fa1f10a4f761b8d2f","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/6606434fa1f10a4f761b8d2f/jM2IQN-xFJi-Z2VcUPn93.jpeg","isPro":false,"fullname":"Pengcheng Wang","user":"PengchengW","type":"user"},{"_id":"65b38b6385b6c214481d016e","avatarUrl":"/avatars/2965359c4b2f571f7f1ac4482de615ec.svg","isPro":false,"fullname":"Haojie","user":"yehaojie1","type":"user"},{"_id":"63954c8efb70088a0a3b272c","avatarUrl":"/avatars/749e101da51067b3569c17cf3acad9a8.svg","isPro":false,"fullname":"Jiacheng Zhang","user":"Joyceool","type":"user"},{"_id":"686eb7537730fef2bdc9e3b3","avatarUrl":"/avatars/607501566fdd964bb4199e68865f9c62.svg","isPro":false,"fullname":"Ziheng","user":"Libero0809","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":0,"organization":{"_id":"65448bef5b5d9185ba3202b9","name":"UIUC-CS","fullname":"University of Illinois at Urbana-Champaign","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/65448b21fcb96b8b48733729/ycqcXFayMTTD_KpE37067.jpeg"},"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2605/2605.22642.md"}">
Papers
arxiv:2605.22642
Spreadsheet-RL: Advancing Large Language Model Agents on Realistic Spreadsheet Tasks via Reinforcement Learning
Authors:
,
,
,
,
,
,
,
,
,
,
,
Abstract
Spreadsheet-RL is a reinforcement learning framework that trains specialized spreadsheet agents in realistic Excel environments, improving AI agent performance on both general and domain-specific spreadsheet tasks through automated data collection and domain-specific benchmarks.
AI-generated summary
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.
Community
Spreadsheet-RL is an RL fine-tuning framework and benchmarking environment designed to improve LLM agent performance on complex, multi-step spreadsheet tasks within Microsoft Excel.
Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.
Tap or paste here to upload images
Get this paper in your agent:
hf papers read 2605.22642 curl -LsSf https://hf.co/cli/install.sh | bash Models citing this paper 0
No model linking this paper
Cite arxiv.org/abs/2605.22642 in a model README.md to link it from this page.
Datasets citing this paper 1
Spaces citing this paper 0
No Space linking this paper
Cite arxiv.org/abs/2605.22642 in a Space README.md to link it from this page.
Collections including this paper 0
No Collection including this paper
Add this paper to a collection to link it from this page.
Discussion (0)
Sign in to join the discussion. Free account, 30 seconds — email code or GitHub.
Sign in →No comments yet. Sign in and be the first to say something.