Datagrom AI News Logo

Microsoft’s new rStar-Math technique upgrades small models to outperform OpenAI’s o1-preview at math problems

Microsoft’s new rStar-Math technique upgrades small models to outperform OpenAI’s o1-preview at math problems

January 9, 2025: Microsoft Boosts Small Models with rStar-Math - Microsoft's new rStar-Math technique enhances small language models (SLMs) using Monte Carlo Tree Search (MCTS) and self-evolution for advanced math problem-solving. It outperforms OpenAI's o1-preview model in key benchmarks, notably achieving a 90% accuracy on the MATH benchmark and excelling in the AIME.

This development highlights the potential of efficient, smaller models, offering powerful alternatives without the high costs and environmental impact of larger systems. The code is set for release post-review, expanding access to these innovative methods.

Link to article Share on LinkedIn

Get the Edge in AI – Join Thousands Staying Ahead of the Curve

Weekly insights on AI trends, industry breakthroughs, and exclusive analysis from leading experts.

Only valuable weekly AI insights. No spam – unsubscribe anytime.