<ul><li>A new open-source math model called Light-R1-32B has been introduced and is optimized for solving advanced math problems.</li><li>The 32-billion parameter model surpasses the performance of equivalent open-source models such as DeepSeek-R1-Distill-Llama-70B and DeepSeek-R1-Distill-Qwen-32B on the American Invitational Mathematics Examination (AIME).</li><li>The training of Light-R1-32B was completed in under six hours on 12 Nvidia H800 GPUs, with an estimated cost of only $1000.</li><li>The researchers have released the model's training datasets, training scripts, and evaluation tools, making it transparent and accessible for building math-focused AI models.</li></ul>

New open-source math model Light-R1-32B surpasses equivalent DeepSeek performance with only $1000 in training costs

Discover more