knoveleng/Open-RS3

Open-RS3 by knoveleng is a 1.5 billion parameter language model, distilled from DeepSeek-R1-Distill-Qwen-1.5B, and enhanced with reinforcement learning to significantly improve mathematical and general reasoning capabilities. It achieves notable performance on benchmarks like AIME24 (46.7%) and AMC23 (80%), surpassing larger models like o1-preview in specific reasoning tasks. This model demonstrates a cost-effective approach to boosting reasoning in small LLMs, making it suitable for resource-constrained applications requiring strong analytical skills.