Apriel-Nemotron-15B-Thinker
Abstract
While large language models (LLMs) have achieved remarkable reasoning capabilities across domains like code, math and other enterprise tasks, their significant memory and computational costs often preclude their use in practical enterprise settings. To this end, we introduce Apriel-Nemotron-15B-Thinker, a 15-billion parameter model in the ServiceNow Apriel SLM series that achieves performance against medium sized state-of-the-art models such as o1-mini, QWQ32B, and EXAONE-Deep-32B while maintaining only half the memory footprint of those alternatives. Apriel-Nemotron-15B-Thinker model is trained in a four stage training pipeline including 1) Base Model upscaling, 2) Continual Pre-training 3) Supervised Fine-tuning (SFT) and 4) Reinforcement Learning using GRPO. Comprehensive evaluations across a diverse suite of benchmarks consistently demonstrate that our Apriel-Nemotron-15B-Thinker model matches or exceeds the performance of its 32-billion parameter counterparts, despite being less than half their size.
Cite
@article{arxiv.2508.10948,
title = {Apriel-Nemotron-15B-Thinker},
author = {Shruthan Radhakrishna and Soham Parikh and Gopal Sarda and Anil Turkkan and Quaizar Vohra and Raymond Li and Dhruv Jhamb and Kelechi Ogueji and Aanjaneya Shukla and Oluwanifemi Bamgbose and Toby Liang and Luke Kumar and Oleksiy Ostapenko and Shiva Krishna Reddy Malay and Aman Tiwari and Tara Bogavelli and Vikas Yadav and Jash Mehta and Saloni Mittal and Akshay Kalkunte and Pulkit Pattnaik and Khalil Slimi and Anirudh Sreeram and Jishnu Nair and Akintunde Oladipo and Shashank Maiya and Khyati Mahajan and Rishabh Maheshwary and Masoud Hashemi and Sai Rajeswar Mudumba and Sathwik Tejaswi Madhusudhan and Torsten Scholak and Sebastien Paquet and Sagar Davasam and Srinivas Sunkara},
journal= {arXiv preprint arXiv:2508.10948},
year = {2025}
}