Skywork-OR1-Math-7B: Specialized Mathematical Reasoning

Skywork-OR1-Math-7B is a 7.6 billion parameter model from the Skywork-OR1 series, developed by Skywork. It is specifically engineered for advanced mathematical reasoning, distinguishing itself from general-purpose language models through its specialized training methodology.

Key Capabilities

• Exceptional Mathematical Reasoning: Achieves a score of 69.8 on AIME24 and 52.3 on AIME25, outperforming other models of similar size in these benchmarks.
• Reinforcement Learning Training: Developed using large-scale rule-based reinforcement learning (RL) with meticulously curated datasets and a multi-stage training pipeline.
• Data-driven Optimization: Incorporates model-aware difficulty estimation, offline and online difficulty-based filtering, and rejection sampling to enhance training efficiency and effectiveness.

Good For

• Complex Math Problem Solving: Ideal for applications requiring high accuracy in mathematical reasoning and problem-solving.
• Research in Reasoning Models: Provides a strong baseline and open-source resources (data, code, blog) for further research into open reasoning models.