---
license: mit
---

# math_bench

Evaluation dataset for Qwen3 math reasoning.

## Source
Randomly sampled 1000 problems from `open-r1/OpenR1-Math-220k` (seed=42).

## Columns
| Column | Description |
|---|---|
| `problem` | Math question |
| `answer` | Reference answer |
| `subject` | Problem type (renamed from `problem_type`) |
| `level` | Source dataset (renamed from `source`) |