The AI Mathematical Olympiad (AIMO) on Kaggle has announced its winners, with the top four teams all utilizing the DeepSeekMath-7B model. Numina's model, NuminaMath-7B-TIR, achieved a notable score of 29 out of 50 correct answers, impressing even renowned mathematician Terence Tao. The model, which won the first progress prize, was fine-tuned in two stages: first, learning math through chain-of-thought samples, and second, learning code with tool-integrated reasoning. The model's success highlights the effectiveness of combining large language models with code execution and synthetic data to solve complex mathematical problems. NuminaMath-7B-TIR scored 29/50 on both public and private Kaggle test sets and is now available as an open-source model under the Apache 2.0 license. Constraints of Kaggle (T4 GPUs) required the use of activation aware quantization to maintain model performance.