https://arxivlens.com/PaperView/Details/recursive-language-models-6610-16b3d94b

Executive Summary
Detailed Breakdown
Practical Applications

\n","updatedAt":"2026-01-06T22:50:41.180Z","author":{"_id":"65243980050781c16f234f1f","avatarUrl":"/avatars/743a009681d5d554c27e04300db9f267.svg","fullname":"Avi","name":"avahal","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":2,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.7436873912811279},"editors":["avahal"],"editorAvatarUrls":["/avatars/743a009681d5d554c27e04300db9f267.svg"],"reactions":[],"isReport":false}},{"id":"695db8b0af2e5efffec93c11","author":{"_id":"63d3e0e8ff1384ce6c5dd17d","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1674830754237-63d3e0e8ff1384ce6c5dd17d.jpeg","fullname":"Librarian Bot (Bot)","name":"librarian-bot","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":305,"isUserFollowing":false},"createdAt":"2026-01-07T01:36:48.000Z","type":"comment","data":{"edited":false,"hidden":false,"latest":{"raw":"This is an automated message from the [Librarian Bot](https://huggingface.co/librarian-bots). I found the following papers similar to this paper. \n\nThe following papers were recommended by the Semantic Scholar API \n\n* [Behavior-Equivalent Token: Single-Token Replacement for Long Prompts in LLMs](https://huggingface.co/papers/2511.23271) (2025)\n* [Sentence-Anchored Gist Compression for Long-Context LLMs](https://huggingface.co/papers/2511.08128) (2025)\n* [Reasoning About Intent for Ambiguous Requests](https://huggingface.co/papers/2511.10453) (2025)\n* [Structured Prompting Enables More Robust Evaluation of Language Models](https://huggingface.co/papers/2511.20836) (2025)\n* [BRAID: Bounded Reasoning for Autonomous Inference and Decisions](https://huggingface.co/papers/2512.15959) (2025)\n* [Towards Small Language Models for Security Query Generation in SOC Workflows](https://huggingface.co/papers/2512.06660) (2025)\n* [Effect of Document Packing on the Latent Multi-Hop Reasoning Capabilities of Large Language Models](https://huggingface.co/papers/2512.14427) (2025)\n\n\n Please give a thumbs up to this comment if you found it helpful!\n\n If you want recommendations for any Paper on Hugging Face checkout [this](https://huggingface.co/spaces/librarian-bots/recommend_similar_papers) Space\n\n You can directly ask Librarian Bot for paper recommendations by tagging it in a comment: `@librarian-bot recommend`","html":"

This is an automated message from the Librarian Bot. I found the following papers similar to this paper.

The following papers were recommended by the Semantic Scholar API

Please give a thumbs up to this comment if you found it helpful!

If you want recommendations for any Paper on Hugging Face checkout this Space

You can directly ask Librarian Bot for paper recommendations by tagging it in a comment: \n\n@librarian-bot\n\t recommend

\n","updatedAt":"2026-01-07T01:36:48.863Z","author":{"_id":"63d3e0e8ff1384ce6c5dd17d","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1674830754237-63d3e0e8ff1384ce6c5dd17d.jpeg","fullname":"Librarian Bot (Bot)","name":"librarian-bot","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":305,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.6958038806915283},"editors":["librarian-bot"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/1674830754237-63d3e0e8ff1384ce6c5dd17d.jpeg"],"reactions":[],"isReport":false}},{"id":"695e586a7ad0bc58078aba01","author":{"_id":"695e57f26dcb58dc4a10bedb","avatarUrl":"/avatars/78d9c0917658b126c5074772c2c47776.svg","fullname":"Brett Earley","name":"klietus","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"isUserFollowing":false},"createdAt":"2026-01-07T12:58:18.000Z","type":"comment","data":{"edited":false,"hidden":false,"latest":{"raw":"Not bad. Almost there. If you make it a graph though, of symbolic concepts you can create something much larger.\n\nhttps://signal-zero.ai","html":"

Not bad. Almost there. If you make it a graph though, of symbolic concepts you can create something much larger.

https://signal-zero.ai

\n","updatedAt":"2026-01-07T12:58:18.643Z","author":{"_id":"695e57f26dcb58dc4a10bedb","avatarUrl":"/avatars/78d9c0917658b126c5074772c2c47776.svg","fullname":"Brett Earley","name":"klietus","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.8615755438804626},"editors":["klietus"],"editorAvatarUrls":["/avatars/78d9c0917658b126c5074772c2c47776.svg"],"reactions":[],"isReport":false}},{"id":"695fc7c39ccaaf87b4aac892","author":{"_id":"661ab1f1fa3b144a381fa454","avatarUrl":"/avatars/7e11b350db47f3ae82dc073f24833a02.svg","fullname":"Urro","name":"urroxyz","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"isUserFollowing":false},"createdAt":"2026-01-08T15:05:39.000Z","type":"comment","data":{"edited":false,"hidden":false,"latest":{"raw":"Isn't this just creating a tool call for the model itself?","html":"

Isn't this just creating a tool call for the model itself?

\n","updatedAt":"2026-01-08T15:05:39.879Z","author":{"_id":"661ab1f1fa3b144a381fa454","avatarUrl":"/avatars/7e11b350db47f3ae82dc073f24833a02.svg","fullname":"Urro","name":"urroxyz","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.924056887626648},"editors":["urroxyz"],"editorAvatarUrls":["/avatars/7e11b350db47f3ae82dc073f24833a02.svg"],"reactions":[],"isReport":false}},{"id":"696707da7b115954f1bf92cf","author":{"_id":"65d9fc2a0e6ad24551d87a1e","avatarUrl":"/avatars/3aedb9522cc3cd08349d654f523fd792.svg","fullname":"Grant Singleton","name":"grantsing","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":4,"isUserFollowing":false},"createdAt":"2026-01-14T03:04:58.000Z","type":"comment","data":{"edited":false,"hidden":false,"latest":{"raw":"arXiv explained breakdown of this paper 👉 https://arxivexplained.com/papers/recursive-language-models\n","html":"

arXiv explained breakdown of this paper 👉 https://arxivexplained.com/papers/recursive-language-models

\n","updatedAt":"2026-01-14T03:04:58.749Z","author":{"_id":"65d9fc2a0e6ad24551d87a1e","avatarUrl":"/avatars/3aedb9522cc3cd08349d654f523fd792.svg","fullname":"Grant Singleton","name":"grantsing","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":4,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.7509856224060059},"editors":["grantsing"],"editorAvatarUrls":["/avatars/3aedb9522cc3cd08349d654f523fd792.svg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2512.24601","authors":[{"_id":"695b6f84832867f253525e5a","name":"Alex L. Zhang","hidden":false},{"_id":"695b6f84832867f253525e5b","name":"Tim Kraska","hidden":false},{"_id":"695b6f84832867f253525e5c","name":"Omar Khattab","hidden":false}],"publishedAt":"2025-12-31T03:43:41.000Z","submittedOnDailyAt":"2026-01-06T04:56:19.174Z","title":"Recursive Language Models","submittedOnDailyBy":{"_id":"64b8e82aa62c52b252c827fa","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/64b8e82aa62c52b252c827fa/Jyk5PHMXCaRlmWy4mT3Bt.jpeg","isPro":true,"fullname":"Rajkumar rawal","user":"rajkumarrawal","type":"user"},"summary":"We study allowing large language models (LLMs) to process arbitrarily long prompts through the lens of inference-time scaling. We propose Recursive Language Models (RLMs), a general inference strategy that treats long prompts as part of an external environment and allows the LLM to programmatically examine, decompose, and recursively call itself over snippets of the prompt. We find that RLMs successfully handle inputs up to two orders of magnitude beyond model context windows and, even for shorter prompts, dramatically outperform the quality of base LLMs and common long-context scaffolds across four diverse long-context tasks, while having comparable (or cheaper) cost per query.","upvotes":68,"discussionId":"695b6f84832867f253525e5d","projectPage":"https://alexzhang13.github.io/blog/2025/rlm/","githubRepo":"https://github.com/alexzhang13/rlm/tree/main","githubRepoAddedBy":"user","githubStars":1352,"organization":{"_id":"63728bde14d543d507ae970d","name":"MIT","fullname":"Massachusetts Institute of Technology","avatar":"https://cdn-uploads.huggingface.co/production/uploads/noauth/S90qoeEJeEYaYf-c7Zs8g.png"}},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"64834b399b352597e41816ac","avatarUrl":"/avatars/63d9d123bffa90f43186a0bdc4455cbd.svg","isPro":false,"fullname":"Shaobai Jiang","user":"shaobaij","type":"user"},{"_id":"64b8e82aa62c52b252c827fa","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/64b8e82aa62c52b252c827fa/Jyk5PHMXCaRlmWy4mT3Bt.jpeg","isPro":true,"fullname":"Rajkumar rawal","user":"rajkumarrawal","type":"user"},{"_id":"62d8d034e2b34bed7442bc1a","avatarUrl":"/avatars/d27698e35668e4b0de00f9fe83f2b8aa.svg","isPro":false,"fullname":"John Posada","user":"elchulito89","type":"user"},{"_id":"60eeedbf50b60c406afc1291","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1649111275459-60eeedbf50b60c406afc1291.png","isPro":false,"fullname":"Samuel Arcadinho","user":"SSamDav","type":"user"},{"_id":"684d57f26e04c265777ead3f","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/cuOj-bQqukSZreXgUJlfm.png","isPro":false,"fullname":"Joakim Lee","user":"Reinforcement4All","type":"user"},{"_id":"6323f399462470712720c155","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/6323f399462470712720c155/SWsMNa7vETUSrOt9Qf-oe.png","isPro":false,"fullname":"Yinxu Pan","user":"cppowboy","type":"user"},{"_id":"64706424d9360cd9d8e5b0dc","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/64706424d9360cd9d8e5b0dc/NcGoYhmm20uZXxP73yx-P.jpeg","isPro":false,"fullname":"Alex","user":"M0nteCarl0","type":"user"},{"_id":"62618a61aee4bea02140d7d0","avatarUrl":"/avatars/9fd4813500df7bb8d64a15efd869a9d0.svg","isPro":false,"fullname":"Shivaen Ramshetty","user":"shivr","type":"user"},{"_id":"62ea79dd01ed9b0e8f61ccd3","avatarUrl":"/avatars/70af83e0e267be39fcd5f23b85e2dafa.svg","isPro":false,"fullname":"Chengsong Huang","user":"ChengsongHuang","type":"user"},{"_id":"67bcaba1608ec2bdb922e8ec","avatarUrl":"/avatars/ed6a3495586b3aaadd298a3d2d6cda59.svg","isPro":false,"fullname":"Alex Zhang","user":"a1zhang","type":"user"},{"_id":"67d32ef182c5e791d544701a","avatarUrl":"/avatars/96f34732e37adced8d2cb92d23eee42a.svg","isPro":false,"fullname":"Umamahesh Janapareddi","user":"Umamahesh-J","type":"user"},{"_id":"645ac130c4acfcf664028c73","avatarUrl":"/avatars/92c2c58e1d3506cbed9dc9d59f79f363.svg","isPro":false,"fullname":"John Giles III","user":"maxnumbers","type":"user"}],"acceptLanguages":["*"],"dailyPaperRank":3,"organization":{"_id":"63728bde14d543d507ae970d","name":"MIT","fullname":"Massachusetts Institute of Technology","avatar":"https://cdn-uploads.huggingface.co/production/uploads/noauth/S90qoeEJeEYaYf-c7Zs8g.png"}}">

Papers

arxiv:2512.24601

Recursive Language Models

Published on Dec 31, 2025

· Submitted by

Rajkumar rawal on Jan 6

#3 Paper of the day

Massachusetts Institute of Technology

Upvote

Authors:

Abstract

We study allowing large language models (LLMs) to process arbitrarily long prompts through the lens of inference-time scaling. We propose Recursive Language Models (RLMs), a general inference strategy that treats long prompts as part of an external environment and allows the LLM to programmatically examine, decompose, and recursively call itself over snippets of the prompt. We find that RLMs successfully handle inputs up to two orders of magnitude beyond model context windows and, even for shorter prompts, dramatically outperform the quality of base LLMs and common long-context scaffolds across four diverse long-context tasks, while having comparable (or cheaper) cost per query.

View arXiv page View PDF Project page GitHub 1.35k Add to collection

Community

rajkumarrawal

Paper submitter 13 days ago

•

edited 13 days ago

Study allowing large language models (LLMs) to process arbitrarily long prompts through the lens of inference-time scaling. They propose Recursive Language Models (RLMs), a general inference strategy that treats long prompts as part of an external environment and allows the LLM to programmatically examine, decompose, and recursively call itself over snippets of the prompt. They find that RLMs successfully handle inputs up to two orders of magnitude beyond model context windows and, even for shorter prompts, dramatically outperform the quality of base LLMs and common long-context scaffolds across four diverse long-context tasks, while having comparable (or cheaper) cost per query.

Some of the observations they found are :-
-- LLMs interacting with their own prompts as objects.

-- In their approach, a prompt isn’t “run” directly, instead it’s stored as a variable in an external Python REPL, and the language model writes code to inspect /slice/ decompose that long string, observes execution outputs, and then constructs sub-tasks where it recursively invokes an LLM on just the relevant snippets. Stitching the result together when the recursive process ends. So it can solve 10M+ token tasks with far less “context rot” and often lower cost than summarization/RAG, turning long-context scaling into an inference-time algorithm rather than just a bigger context window.

-- The ability to search the Prompt is what enables handling long context inputs, sub calls help handle information dense inputs.

-- Inference cost of RLMs remain comparable to a base model call but are high variance because it can keep making sub-calls or iterate if it can't solve the problem initially.

-- The key insight is that long prompts should not be fed into the LLM directly, but should instead be treated as part of the environment that the LLM can search, read and interact with as needed for the task.