Authors
Arash Ebrahim Abtahizadeh, Muna Aghamelu, Adil Ahmed, Juan C. Angeles Acuna, Alexis Audran-Reiss, Asaf Avrahamy, Yoram Bachrach, Can Balioglu, Amar Budhiraja, Onur Çelebi, Gaurav Chaurasia, Joy Chen, Lynn Cherif, Lucia Cipolina-Kun, Gal Cohen, Timothée Darcet, Beto de Paola, Digant Desai, Sam Devlin, Derek Dunfield, Ivan Evtimov, Maryam Fazel-Zarandi, Rob Fergus, Jean-Christophe Gagnon-Audet, Grant Gardner, Bhavul Gauri, Paris Giampouras, Sujan Kumar Gonugondla, Antoine Grosnit, Young Han, Michael Hassid, Monuril Islam, Masoud Jalili Sabet, Songhao Jia, Narine Kokhlikyan, Svetlana Kurtsman, Sandra Lefdal, Chee Hau Leow, Noam Levi, Mike Lewis, Sanae Lotfi, Xiyang Luo, Despoina Magka, Saeed Mahloujifar, Neal Mangaokar, Ishita Mediratta, Alexander H. Miller, Hela Momand, Saba Nazir, Kelvin Niu, Mattia Opper, Parth Pathak, Alberto Pepe, Danielle Pintz, Mikhail Plekhanov, Sidhartha Reddy Potu, Anton Protopopov, Shengyi Qian, Shashank Rajput, Joseph Rocca, Scott Roy, Luca Sbordone, Amin Shakeri, Tatiana Shavrina, Michael Shvartsman, Ruan Silva, Anya Sims, Eric Michael Smith, Hugo Touvron, Mabel Ubong, Isabel Urrego, Yanzheng Xiang, Zhiyuan Yao, Alex Yarmula, He Ye, Jade Yu, Luke Zettlemoyer, Arman Zharmagambetov, Perpetua Fathima Ahlijah, Tremo Ahmed, Enrique Alfonseca, Pedram Bashiri, Nikolay Bashlykov, Aaron Berman, Rohith Bhatia, Siddharth Bhattacharya, Prashant Bhende, Chloe Bi, Shuchao Bi, Clement Anthonioz Blanc, Allen Bolourchi, Andrea Caciolai, Erica Cai, Xiang Tony Cao, Yue Cao, Andrew Caples, Jinzhao Chang, Lele Chen, Moya Chen, Weijie Chen, Wenhu Chen, Xiaobai Chen, Yinyin Chen, Zhengxing Chen, Wei Cheng, Pavel Chikulaev, Edward Chou, Sy Choudhury, Brian Clerkin, Marta R. Costa-jussà, Curtis Davis, Myra Deng, Dujian Ding, Hantian Ding, Shuangrui Ding, Jesse Dodge, Shweta Doshi, Shirin Eghtesadi, Faith Eischen, Athyu Eleti, Alex Fabbri, Han Fang, Paula Farani, Aiman Farooq, Christoph Feichtenhofer, Darius Feher, Amanda Felix, Carl Chengyan Fu, Sean Gahagan, Yifan Gao, Itai Gat, Akashdeep Goel, Ahuva Goldstand, Bo He, David Heineman, Dylan Herman, Irene Hjelt, Joonas Hjelt, Alexander Holz, Rui Hou, Annie Hu, Yuan-Ting Hu, Bernie Huang, Jing Huang, Kun Huang, Pere-Lluís Huguet Cabot, Ashley Hunter, Srinivasan Iyer, Somya Jain, Michael Jiang, Song Jiang, Wenwen Jiang, Lin Jin, Jacob Kahn, Melanie Kambadur, Yingxin Kang, Justine Kao, Stephane Kasriel, Hossein Kavianihamedani, Ameya Khandekar, Yash Khare, Gefen Kohavi, Kavya Kopparapu, Praveen Krishnan, Venu Kuchibhotla, Ananya Kumar, Leona Lan, Jack Lanchantin, Hunter Lang, Nikolay Pavlovich Laptev, Yuanzhi Li, Yunlu Li, Asad Liaqat, Johnson Lin, Runwei Lin, Zhichao Lin, Shaun Lindsay, Yaron Lipman, Frederick Liu, Mengchen Liu, Qiang Liu, Yinghai Lu, Yujie Lu, Christopher Luc, Faye Ma, Pingchuan Ma, Tengyu Ma, Wufei Ma, Xiaodong Ma, Damir Maksyutov, Josh Malone, Udeep Manchanda, Thomas Mann, Kalyani Marathe, Roberto-Rafael Maura-Rivero, Ruslan Mavlyutov, Thomas Mesnard, Julian Michael, Sagar Miglani, Todor Mihaylov, Naureen Mitha, Archi Mitra, Leo Mok, Naila Murray, Muthukumaravel Muthuraman, Mergen Nachin, Nese Demir Nalci, Saaketh Narayan, Alyssa Newcomb, Peter Ney, Joycelyn Ng, Devin Norder, Maxwell Nye, Rajvardhan Oak, Cody Ohlsen, Artidoro Pagnoni, Yi Pan, Nilesh Pandey, Guan Pang, Richard Yuanzhe Pang, Mahesh Pasupuleti, Parth Patel, Yash Patel, Vladan Petrovic, Norman Ponte, Yundi Qian, Zengyi Qin, Jack Rae, Emmanuel Ramirez, Ramya Raghavendra, Nikhila Ravi, Hongyu Ren, Riandy Riandy, Harrison Rudolph, Kalyan Saladi, Eduardo Sánchez, Karthik Abinav Sankararaman, Aman Shankar, Paresh Shenoy, Alana Shine, Hamid Shojanazeri, Iva Simon-Bubalo, Mitesh Kumar Singh, Tiffany Siqi, Pawan Sisodiya, Ali Sol, Sten Sootla, Leslie St. Pierre, DiJia Su, Suraj Subramanian, Brian Sun, Tony Sun, Zhiqing Sun, Natacha Supper, Jason Talpalar, Ellen Xiaoqing Tan, Liang Tan, Jean-Christophe Testud, Parth Thakkar, Joseph Tighe, Ty Toledano, T Tse, Igor Tufanov, Alex Vaughan, Josh Walters, Alex Wang, Bethany Wang, Meng Wang, Rocky Wang, Weiyue Wang, Xiaolan Wang, Xin Wang, Zijian Wang, Bram Wasti, Quentin Wellens, Yeming Wen, Mary Williamson, Yuncheng Wu, Fei Xia, Tian Xie, David Xu, Jacob Xu, Puxin Xu, Zheng Xu, Shen Yan, Megan Yang, Yining Yang, Nick Yannopoulos, Kevin Yao, Michihiro Yasunaga, Haoxuan You, Michael You, Dingli Yu, Tianhe Yu, Summer Yue, Xiang Yue, Shuwei Zeng, Hanwen Zha, Chen Zhang, Hongming Zhang, Kaiyu Zhang, Meng Zhang, Mingyang Zhang, Yi Zhang, Shengjia Zhao, Chao Zhou, Claire Zhou, Shuyan Zhou, Yanjun Zhou, Yaxuan Zhou, Rosalie Zhu, Yun Zhu
Affiliations
Meta
Categories
Method / Agentic Models / Integrated reasoning and tool usage, Application / On-Device AI / Consumer device local execution, Evaluation / Model Reliability / Fault recovery in causal LLMs
License
CC BY 4.0
Abstract Overview
Muse Glimmer is an open-source 30B-parameter dense causal language model equipped with a dedicated perception encoder, designed for autonomous agentic workflows running locally on consumer hardware. The model integrates multi-step reasoning, tool execution, and failure recovery within a single system that handles interleaved text and image inputs across approximately 140 languages with context lengths of 128K+ tokens. Built to fit within a 32 GB VRAM ceiling, its training recipe combines teacher logit distillation from Muse Spark with agentic mid-training, supervised fine-tuning, on-policy distillation, and reinforcement learning. The release also incorporates deployment-oriented enhancements, such as post-training 4-bit quantization, speculative decoding via a DFlash head, and train-time mitigations for privacy and adversarial robustness.
Novelty
The work introduces a purpose-built, open-weight 30B dense agentic model engineered specifically for on-device consumer hardware rather than cloud deployment. It unifies multimodal perception, configurable reasoning depth, agentic tool-use alignment, privacy-preserving information flow training, and efficient inference artifacts including an evolutionary mixed-precision quantization scheme and a speculative decoding head in a single open release.
Results
Muse Glimmer achieves leading results among models in its size class on multiple agentic and tool-use benchmarks, scoring 75.5 on MCPAtlas, 43.3 on GAIA2, and 51.2 on SWE-Bench Pro while remaining competitive with or exceeding Gemma4-31B and Qwen3.6-27B. Quantization evaluations across 15 downstream tasks show minimal quality loss, with the EvoPress profile attaining 75.6% average accuracy compared to 75.7% in full BF16. The model also exhibits strong long-context and multilingual capabilities, recording 65.1 on Beam 128K and 60.4 on Omnilingual Gaia2.
Key Points
- Muse Glimmer is a 30B dense causal language model with a frozen vision perception encoder and 128K+ context length, architecturally designed to execute within a 32 GB VRAM envelope on consumer hardware.
- The training pipeline incorporates logit distillation from the frontier Muse Spark model, agentic synthetic universe data, on-policy distillation, and multi-objective reinforcement learning for reasoning, tool use, and privacy-aware information flow.
- Evaluations demonstrate state-of-the-art agentic and tool-use performance for its scale, maintaining near-BF16 accuracy when compressed to approximately 4-bit quantization and accelerated via a DFlash speculation head.
References