Qwen2.5-Math: The world's leading open-sourced mathematical LLMs

Qwen's Qwen2.5-Math combines written reasoning with a Python interpreter to solve English and Chinese mathematics problems. The model can generate code for calculations or symbolic operations that are difficult to carry out reliably through text alone. The team's release article describes this as tool-integrated reasoning.

The reported benefit depends on how answers are produced. In the team's evaluation, the 7B instruction-tuned model scored 83.6 with text reasoning and 85.3 with tool-integrated reasoning on MATH, a mathematics test set. Those results support evaluating the model together with its calculation tools. The series is specialized for mathematics; the team advises against using it for unrelated tasks.