https://github.com/stepfun-ai/PaCoRe
Data: https://huggingface.co/datasets/stepfun-ai/PaCoRe-Train-8k
Model: https://huggingface.co/stepfun-ai/PaCoRe-8B

\n","updatedAt":"2026-01-13T03:52:07.648Z","author":{"_id":"625026b7d2d191ac43320c5e","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/625026b7d2d191ac43320c5e/2ExzHlZ-Bk8SQMyBjeY6N.jpeg","fullname":"Jingcheng Hu","name":"reign12","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":24,"isUserFollowing":false}},"numEdits":1,"identifiedLanguage":{"language":"en","probability":0.8166795372962952},"editors":["reign12"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/625026b7d2d191ac43320c5e/2ExzHlZ-Bk8SQMyBjeY6N.jpeg"],"reactions":[],"isReport":false}},{"id":"6966f44d2b50974deada2921","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":318,"isUserFollowing":false},"createdAt":"2026-01-14T01:41:33.000Z","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* [AgentMath: Empowering Mathematical Reasoning for Large Language Models via Tool-Augmented Agent](https://huggingface.co/papers/2512.20745) (2025)\n* [Native Parallel Reasoner: Reasoning in Parallelism via Self-Distilled Reinforcement Learning](https://huggingface.co/papers/2512.07461) (2025)\n* [Falcon-H1R: Pushing the Reasoning Frontiers with a Hybrid Model for Efficient Test-Time Scaling](https://huggingface.co/papers/2601.02346) (2026)\n* [Nanbeige4-3B Technical Report: Exploring the Frontier of Small Language Models](https://huggingface.co/papers/2512.06266) (2025)\n* [Dual-Phase LLM Reasoning: Self-Evolved Mathematical Frameworks](https://huggingface.co/papers/2601.05616) (2026)\n* [CoSineVerifier: Tool-Augmented Answer Verification for Computation-Oriented Scientific Questions](https://huggingface.co/papers/2512.01224) (2025)\n* [Motif-2-12.7B-Reasoning: A Practitioner's Guide to RL Training Recipes](https://huggingface.co/papers/2512.11463) (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":"2026-01-14T01:41:33.758Z","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":318,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.692028284072876},"editors":["librarian-bot"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/1674830754237-63d3e0e8ff1384ce6c5dd17d.jpeg"],"reactions":[],"isReport":false}},{"id":"696b8acfd0d46e96ec3b95aa","author":{"_id":"65243980050781c16f234f1f","avatarUrl":"/avatars/743a009681d5d554c27e04300db9f267.svg","fullname":"Avi","name":"avahal","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":3,"isUserFollowing":false},"createdAt":"2026-01-17T13:12:47.000Z","type":"comment","data":{"edited":false,"hidden":false,"latest":{"raw":"arXivlens breakdown of this paper 👉 https://arxivlens.com/PaperView/Details/pacore-learning-to-scale-test-time-compute-with-parallel-coordinated-reasoning-1362-71d9f506\n\n- Executive Summary\n- Detailed Breakdown\n- Practical Applications","html":"

arXivlens breakdown of this paper 👉 https://arxivlens.com/PaperView/Details/pacore-learning-to-scale-test-time-compute-with-parallel-coordinated-reasoning-1362-71d9f506

Executive Summary
Detailed Breakdown
Practical Applications

\n","updatedAt":"2026-01-17T13:12:47.271Z","author":{"_id":"65243980050781c16f234f1f","avatarUrl":"/avatars/743a009681d5d554c27e04300db9f267.svg","fullname":"Avi","name":"avahal","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":3,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.7720900774002075},"editors":["avahal"],"editorAvatarUrls":["/avatars/743a009681d5d554c27e04300db9f267.svg"],"reactions":[{"reaction":"🚀","users":["reign12"],"count":1}],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2601.05593","authors":[{"_id":"6965b990fc8c4ecc02c7f8df","user":{"_id":"625026b7d2d191ac43320c5e","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/625026b7d2d191ac43320c5e/2ExzHlZ-Bk8SQMyBjeY6N.jpeg","isPro":false,"fullname":"Jingcheng Hu","user":"reign12","type":"user"},"name":"Jingcheng Hu","status":"claimed_verified","statusLastChangedAt":"2026-01-16T10:34:23.867Z","hidden":false},{"_id":"6965b990fc8c4ecc02c7f8e0","name":"Yinmin Zhang","hidden":false},{"_id":"6965b990fc8c4ecc02c7f8e1","name":"Shijie Shang","hidden":false},{"_id":"6965b990fc8c4ecc02c7f8e2","name":"Xiaobo Yang","hidden":false},{"_id":"6965b990fc8c4ecc02c7f8e3","name":"Yue Peng","hidden":false},{"_id":"6965b990fc8c4ecc02c7f8e4","name":"Zhewei Huang","hidden":false},{"_id":"6965b990fc8c4ecc02c7f8e5","name":"Hebin Zhou","hidden":false},{"_id":"6965b990fc8c4ecc02c7f8e6","name":"Xin Wu","hidden":false},{"_id":"6965b990fc8c4ecc02c7f8e7","name":"Jie Cheng","hidden":false},{"_id":"6965b990fc8c4ecc02c7f8e8","name":"Fanqi Wan","hidden":false},{"_id":"6965b990fc8c4ecc02c7f8e9","name":"Xiangwen Kong","hidden":false},{"_id":"6965b990fc8c4ecc02c7f8ea","name":"Chengyuan Yao","hidden":false},{"_id":"6965b990fc8c4ecc02c7f8eb","name":"Kaiwen Yan","hidden":false},{"_id":"6965b990fc8c4ecc02c7f8ec","name":"Ailin Huang","hidden":false},{"_id":"6965b990fc8c4ecc02c7f8ed","name":"Hongyu Zhou","hidden":false},{"_id":"6965b990fc8c4ecc02c7f8ee","name":"Qi Han","hidden":false},{"_id":"6965b990fc8c4ecc02c7f8ef","name":"Zheng Ge","hidden":false},{"_id":"6965b990fc8c4ecc02c7f8f0","name":"Daxin Jiang","hidden":false},{"_id":"6965b990fc8c4ecc02c7f8f1","name":"Xiangyu Zhang","hidden":false},{"_id":"6965b990fc8c4ecc02c7f8f2","name":"Heung-Yeung Shum","hidden":false}],"publishedAt":"2026-01-09T07:24:43.000Z","submittedOnDailyAt":"2026-01-13T00:51:45.124Z","title":"PaCoRe: Learning to Scale Test-Time Compute with Parallel Coordinated Reasoning","submittedOnDailyBy":{"_id":"625026b7d2d191ac43320c5e","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/625026b7d2d191ac43320c5e/2ExzHlZ-Bk8SQMyBjeY6N.jpeg","isPro":false,"fullname":"Jingcheng Hu","user":"reign12","type":"user"},"summary":"We introduce Parallel Coordinated Reasoning (PaCoRe), a training-and-inference framework designed to overcome a central limitation of contemporary language models: their inability to scale test-time compute (TTC) far beyond sequential reasoning under a fixed context window. PaCoRe departs from the traditional sequential paradigm by driving TTC through massive parallel exploration coordinated via a message-passing architecture in multiple rounds. Each round launches many parallel reasoning trajectories, compacts their findings into context-bounded messages, and synthesizes these messages to guide the next round and ultimately produce the final answer. Trained end-to-end with large-scale, outcome-based reinforcement learning, the model masters the synthesis abilities required by PaCoRe and scales to multi-million-token effective TTC without exceeding context limits. The approach yields strong improvements across diverse domains, and notably pushes reasoning beyond frontier systems in mathematics: an 8B model reaches 94.5% on HMMT 2025, surpassing GPT-5's 93.2% by scaling effective TTC to roughly two million tokens. We open-source model checkpoints, training data, and the full inference pipeline to accelerate follow-up work.","upvotes":84,"discussionId":"6965b990fc8c4ecc02c7f8f3","githubRepo":"https://github.com/stepfun-ai/PaCoRe","githubRepoAddedBy":"user","ai_summary":"Parallel Coordinated Reasoning enables large-scale test-time compute scaling beyond sequential reasoning limitations through parallel exploration and message-passing architecture.","ai_keywords":["test-time compute","sequential reasoning","parallel exploration","message-passing architecture","reinforcement learning","multi-million-token","HMMT 2025","GPT-5"],"githubStars":321,"organization":{"_id":"66e43eae9d477f566f937935","name":"stepfun-ai","fullname":"StepFun","avatar":"https://cdn-uploads.huggingface.co/production/uploads/66935cee39002fc0569c2943/Qv8QPbkgoKE3wR4jTzHiy.png"}},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"625026b7d2d191ac43320c5e","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/625026b7d2d191ac43320c5e/2ExzHlZ-Bk8SQMyBjeY6N.jpeg","isPro":false,"fullname":"Jingcheng Hu","user":"reign12","type":"user"},{"_id":"6576c335827f8fd66600a81d","avatarUrl":"/avatars/df5a4dfd2a9fd04b847eb8a3709c44dd.svg","isPro":false,"fullname":"p2oileen","user":"p2oileen","type":"user"},{"_id":"66668c591964b6188ee310c2","avatarUrl":"/avatars/8a8265073dbacbb2c7139b1c8da3e055.svg","isPro":false,"fullname":"Kaiwen Yan","user":"linrany","type":"user"},{"_id":"66aa434994f479755b155896","avatarUrl":"/avatars/1f8cfc423afb35fe29237b87e6f6f4b0.svg","isPro":false,"fullname":"wookimm","user":"wookimm1024","type":"user"},{"_id":"63ae768220176b2d216ac807","avatarUrl":"/avatars/d658b98f469bd12f5bc3bb9c12811780.svg","isPro":false,"fullname":"Zhewei Huang","user":"hzwer","type":"user"},{"_id":"68e3d3038b0b58d3266da523","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/noauth/AU65FydXz9NePJhSztuvR.jpeg","isPro":false,"fullname":"马嘉祺","user":"jiaqima1","type":"user"},{"_id":"67d7f82b81f8b70fecf5cc08","avatarUrl":"/avatars/5e8cf0e495832399e806738a01d5dfef.svg","isPro":false,"fullname":"dong","user":"dongbobo","type":"user"},{"_id":"6721eb52d70f872070c9a0ff","avatarUrl":"/avatars/34239ea757d51dbe22773bd739fc1bab.svg","isPro":false,"fullname":"Sun","user":"Reuben-Sun","type":"user"},{"_id":"664ae39ab5e5f95dc6209365","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/664ae39ab5e5f95dc6209365/8Z9ERYhX6URXh4si6jWGm.jpeg","isPro":false,"fullname":"Yinmin Zhang","user":"YinminZhang","type":"user"},{"_id":"6737605f8e14c110852d0f07","avatarUrl":"/avatars/ee7f1fffd5a8411ac5e87b55b044d4df.svg","isPro":false,"fullname":"Shijie Shang","user":"SuntoryOolong","type":"user"},{"_id":"62ecbffd99112e99c5f7fded","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/62ecbffd99112e99c5f7fded/U6iXAJbpm2vaC5qksEPiH.png","isPro":false,"fullname":"Fanqi Wan","user":"Wanfq","type":"user"},{"_id":"63971c2a3507d82f7976b164","avatarUrl":"/avatars/9387a5b258cd1c80a7a4e71d0fa07994.svg","isPro":false,"fullname":"Jie Cheng","user":"jinachris","type":"user"}],"acceptLanguages":["*"],"dailyPaperRank":3,"organization":{"_id":"66e43eae9d477f566f937935","name":"stepfun-ai","fullname":"StepFun","avatar":"https://cdn-uploads.huggingface.co/production/uploads/66935cee39002fc0569c2943/Qv8QPbkgoKE3wR4jTzHiy.png"}}">

Papers

arxiv:2601.05593

PaCoRe: Learning to Scale Test-Time Compute with Parallel Coordinated Reasoning

Published on Jan 9

· Submitted by

Jingcheng Hu on Jan 13

#3 Paper of the day

StepFun

Upvote

Authors:

Jingcheng Hu ,

Abstract

Parallel Coordinated Reasoning enables large-scale test-time compute scaling beyond sequential reasoning limitations through parallel exploration and message-passing architecture.

AI-generated summary

We introduce Parallel Coordinated Reasoning (PaCoRe), a training-and-inference framework designed to overcome a central limitation of contemporary language models: their inability to scale test-time compute (TTC) far beyond sequential reasoning under a fixed context window. PaCoRe departs from the traditional sequential paradigm by driving TTC through massive parallel exploration coordinated via a message-passing architecture in multiple rounds. Each round launches many parallel reasoning trajectories, compacts their findings into context-bounded messages, and synthesizes these messages to guide the next round and ultimately produce the final answer. Trained end-to-end with large-scale, outcome-based reinforcement learning, the model masters the synthesis abilities required by PaCoRe and scales to multi-million-token effective TTC without exceeding context limits. The approach yields strong improvements across diverse domains, and notably pushes reasoning beyond frontier systems in mathematics: an 8B model reaches 94.5% on HMMT 2025, surpassing GPT-5's 93.2% by scaling effective TTC to roughly two million tokens. We open-source model checkpoints, training data, and the full inference pipeline to accelerate follow-up work.

View arXiv page View PDF GitHub 321 Add to collection

Community

reign12

Paper author Paper submitter Jan 13

•

edited Jan 13

🎉 Introducing Parallel Coordinated Reasoning (PaCoRe)
📈 An 8B model beats GPT-5 on HMMT25 by unlocking parallel thinking for test-time scaling!
📂 Open-source deep think: data + model + inference code!
🆓 MIT-licensed — use it however you want

🔍Key findings:

Message Passing Unlocks Scaling
Without compaction, performance flatlines at the context limit. PaCoRe breaks the memory barrier and lets reasoning scale freely.
Breadth > Depth
All compute is not equal. Coordinated parallel reasoning delivers far higher returns than extending a single chain.
Data as a Force Multiplier
The PaCoRe corpus provides exceptionally valuable supervision— even baseline models see substantial gains when trained on it.

🔗 Links:
GitHub: https://github.com/stepfun-ai/PaCoRe
Data: https://huggingface.co/datasets/stepfun-ai/PaCoRe-Train-8k
Model: https://huggingface.co/stepfun-ai/PaCoRe-8B