| Title: |
Nemotron 3 Nano: Open, Efficient Mixture-of-Experts Hybrid Mamba-Transformer Model for Agentic Reasoning |
| Authors: |
NVIDIA; Blakeman, Aaron; Grattafiori, Aaron; Basant, Aarti; Gupta, Abhibha; Khattar, Abhinav; Renduchintala, Adi; Vavre, Aditya; Shukla, Akanksha; Bercovich, Akhiad; Ficek, Aleksander; Shaposhnikov, Aleksandr; Kondratenko, Alex; Bukharin, Alexander; Milesi, Alexandre; Taghibakhshi, Ali; Liu, Alisa; Barton, Amelia; Mahabaleshwarkar, Ameya Sunil; Klein, Amir; Zuker, Amit; Geifman, Amnon; Shen, Amy; Bhiwandiwalla, Anahita; Tao, Andrew; Guan, Ann; Mandarwal, Anubhav; Mehta, Arham; Aithal, Ashwath; Poojary, Ashwin; Ahamed, Asif; Thekkumpate, Asma Kuriparambil; Dattagupta, Ayush; Zhu, Banghua; Sadeghi, Bardiya; Simkin, Barnaby; Lanir, Ben; Schifferer, Benedikt; Nushi, Besmira; Kartal, Bilal; Rouhani, Bita Darvish; Ginsburg, Boris; Norick, Brandon; Soubasis, Brandon; Kisacanin, Branislav; Yu, Brian; Catanzaro, Bryan; del Mundo, Carlo; Hwang, Chantal; Wang, Charles; Hsieh, Cheng-Ping; Zhang, Chenghao; Yu, Chenhan; Mungekar, Chetan; Patel, Chintan; Alexiuk, Chris; Parisien, Christopher; Neale, Collin; Mosk-Aoyama, Damon; Su, Dan; Corneil, Dane; Afrimi, Daniel; Rohrer, Daniel; Serebrenik, Daniel; Gitman, Daria; Levy, Daria; Stosic, Darko; Mosallanezhad, David; Narayanan, Deepak; Nathawani, Dhruv; Rekesh, Dima; Yared, Dina; Kakwani, Divyanshu; Ahn, Dong; Riach, Duncan; Stosic, Dusan; Minasyan, Edgar; Lin, Edward; Long, Eileen; Long, Eileen Peters; Lantz, Elena; Evans, Ellie; Ning, Elliott; Chung, Eric; Harper, Eric; Tramel, Eric; Galinkin, Erick; Pounds, Erik; Briones, Evan; Bakhturina, Evelina; Ladhak, Faisal; Wang, Fay; Jia, Fei; Soares, Felipe; Chen, Feng; Galko, Ferenc; Siino, Frankie; Agam, Gal Hubara; Ajjanagadde, Ganesh; Bhatt, Gantavya; Prasad, Gargi; Armstrong, George; Shen, Gerald; Batmaz, Gorkem; Nalbandyan, Grigor; Qian, Haifeng; Sharma, Harsh; Ross, Hayley; Ngo, Helen; Sahota, Herman; Wang, Hexin; Soni, Himanshu; Upadhyay, Hiren; Mao, Huizi; Nguyen, Huy C; Nguyen, Huy Q; Cunningham, Iain; Shahaf, Ido; Gitman, Igor; Loshchilov, Ilya; Moshkov, Ivan; Putterman, Izzy; Kautz, Jan; Scowcroft, Jane Polak; Casper, Jared; Mitra, Jatin; Glick, Jeffrey; Chen, Jenny; Oliver, Jesse; Zhang, Jian; Zeng, Jiaqi; Lou, Jie; Zhang, Jimmy; Huang, Jining; Conway, Joey; Guman, Joey; Kamalu, John; Greco, Johnny; Cohen, Jonathan; Jennings, Joseph; Daw, Joyjit; Vialard, Julien Veron; Yi, Junkeun; Parmar, Jupinder; Xu, Kai; Zhu, Kan; Briski, Kari; Cheung, Katherine; Luna, Katherine; Santhanam, Keshav; Shih, Kevin; Kong, Kezhi; Bhardwaj, Khushi; Puvvada, Krishna C.; Pawelec, Krzysztof; Anik, Kumar; McAfee, Lawrence; Sleiman, Laya; Derczynski, Leon; Ding, Li; Liebenwein, Lucas; Vega, Luis; Grover, Maanu; Van Segbroeck, Maarten; de Melo, Maer Rodrigues; Sreedhar, Makesh Narsimhan; Kilaru, Manoj; Ashkenazi, Maor; Romeijn, Marc; Cai, Mark; Kliegl, Markus; Moosaei, Maryam; Novikov, Matvei; Samadi, Mehrzad; Corpuz, Melissa; Wang, Mengru; Price, Meredith; Boone, Michael; Evans, Michael; Martinez, Miguel; Chrzanowski, Mike; Shoeybi, Mohammad; Patwary, Mostofa; Mulepati, Nabin; Hereth, Natalie; Assaf, Nave; Habibi, Negar; Zmora, Neta; Haber, Netanel; Sessions, Nicola; Bhatia, Nidhi; Jukar, Nikhil; Pope, Nikki; Ludwig, Nikolai; Tajbakhsh, Nima; Juluru, Nirmal; Hrinchuk, Oleksii; Kuchaiev, Oleksii; Delalleau, Olivier; Olabiyi, Oluwatobi; Argov, Omer Ullman; Xie, Ouye; Chadha, Parth; Shamis, Pasha; Molchanov, Pavlo; Morkisz, Pawel; Dykas, Peter; Jin, Peter; Xu, Pinky; Januszewski, Piotr; Thombre, Pranav Prashant; Varshney, Prasoon; Gundecha, Pritam; Miao, Qing; Mahabadi, Rabeeh Karimi; El-Yaniv, Ran; Zilberstein, Ran; Shafipour, Rasoul; Harang, Rich; Izzo, Rick; Shahbazyan, Rima; Garg, Rishabh; Borkar, Ritika; Gala, Ritu; Islam, Riyad; Waleffe, Roger; Watve, Rohit; Koren, Roi; Zhang, Ruoxi; Hewett, Russell J.; Prenger, Ryan; Timbrook, Ryan; Mahdavi, Sadegh; Modi, Sahil; Kriman, Samuel; Kariyappa, Sanjay; Satheesh, Sanjeev; Kaji, Saori; Pasumarthi, Satish; Narentharen, Sean; Narenthiran, Sean; Bak, Seonmyeong; Kashirsky, Sergey; Poulos, Seth; Mor, Shahar; Ramasamy, Shanmugam; Acharya, Shantanu; Ghosh, Shaona; Sreenivas, Sharath Turuvekere; Thomas, Shelby; Fan, Shiqing; Gopal, Shreya; Prabhumoye, Shrimai; Pachori, Shubham; Toshniwal, Shubham; Ding, Shuoyang; Singh, Siddharth; Sun, Simeng; Ithape, Smita; Majumdar, Somshubra; Singhal, Soumye; Alborghetti, Stefania; Ge, Stephen; Devare, Sugam Dipak; Barua, Sumeet Kumar; Panguluri, Suseella; Gupta, Suyog; Priyadarshi, Sweta; Akter, Syeda Nahida; Bui, Tan; Ene, Teodor-Dumitru; Kong, Terry; Do, Thanh; Blankevoort, Tijmen; Balough, Tom; Asida, Tomer; Natan, Tomer Bar; Konuk, Tugrul; Vashishth, Twinkle; Karpas, Udi; De, Ushnish; Noorozi, Vahid; Noroozi, Vahid; Srinivasan, Venkat; Elango, Venmugil; Korthikanti, Vijay; Kurin, Vitaly; Lavrukhin, Vitaly; Jiang, Wanli; Ahmad, Wasi Uddin; Du, Wei; Ping, Wei; Zhou, Wenfei; Jennings, Will; Zhang, William; Prazuch, Wojciech; Ren, Xiaowei; Karnati, Yashaswi; Choi, Yejin; Meyer, Yev; Wu, Yi-Fu; Zhang, Yian; Lin, Ying; Geifman, Yonatan; Fu, Yonggan; Subara, Yoshi; Suhara, Yoshi; Gao, Yubo; Moshe, Zach; Dong, Zhen; Liu, Zihan; Chen, Zijia; Yan, Zijie |
| Publication Year: |
2025 |
| Collection: |
ArXiv.org (Cornell University Library) |
| Subject Terms: |
Computation and Language; Artificial Intelligence; Machine Learning |
| Description: |
We present Nemotron 3 Nano 30B-A3B, a Mixture-of-Experts hybrid Mamba-Transformer language model. Nemotron 3 Nano was pretrained on 25 trillion text tokens, including more than 3 trillion new unique tokens over Nemotron 2, followed by supervised fine tuning and large-scale RL on diverse environments. Nemotron 3 Nano achieves better accuracy than our previous generation Nemotron 2 Nano while activating less than half of the parameters per forward pass. It achieves up to 3.3x higher inference throughput than similarly-sized open models like GPT-OSS-20B and Qwen3-30B-A3B-Thinking-2507, while also being more accurate on popular benchmarks. Nemotron 3 Nano demonstrates enhanced agentic, reasoning, and chat abilities and supports context lengths up to 1M tokens. We release both our pretrained Nemotron 3 Nano 30B-A3B Base and post-trained Nemotron 3 Nano 30B-A3B checkpoints on Hugging Face. |
| Document Type: |
text |
| Language: |
unknown |
| Relation: |
http://arxiv.org/abs/2512.20848 |
| Availability: |
http://arxiv.org/abs/2512.20848 |
| Accession Number: |
edsbas.D88216EC |
| Database: |
BASE |