Microsoft AI has introduced rStar-Math, a system designed to enhance the math reasoning capabilities of small language models (LLMs). With a model size of just 7 billion parameters, rStar-Math demonstrates performance that rivals and occasionally surpasses larger models like OpenAI's o1-preview. The system employs a 'deep thinking' approach, using Monte Carlo Tree Search (MCTS) to explore multiple solution paths and select the most effective one. Through 4 rounds of self-evolution with millions of synthesized solutions for 747k math problems, rStar-Math has shown significant improvements on the MATH benchmark, boosting Qwen2.5-Math-7B from 58.8% to 90.0% and Phi3-mini-3.8B from 41.4% to 86.4%, outperforming o1-preview by 4.5% and 0.9% respectively. Additionally, rStar-Math has achieved notable success on the USA Math Olympiad, solving an average of 53.3% of problems, or 8/15 problems, placing it among the top 20% of high school math students.