nvidia/OpenMath-Nemotron-7B

The OpenMath-Nemotron-7B is a 7.6 billion parameter transformer decoder-only language model developed by NVIDIA, fine-tuned from Qwen/Qwen2.5-Math-7B. It is specifically optimized for advanced mathematical reasoning tasks, achieving state-of-the-art results on benchmarks like AIME24, AIME25, HMMT-24-25, and HLE-Math. This model supports a context length of up to 131,072 tokens and is designed for research in mathematical problem-solving.