{"data":{"kind":"file","path":"README.md","version_id":"i0v6ajdl43utdk55ode4qq7u","entry":{"name":"README.md","path":"README.md","is_directory":false,"size":600,"modified_at":"2026-09-13T04:06:06.934000","content_hash":"65b69abbc5b7c27226ad63c3ed6c6d509a314acc332ae446706e9c15f52a013c"},"entries":[],"content":"# arxivmath\n\nMathArena research-level math tasks extracted from arXiv papers, each with a verifiable final answer. The taskset ships only the tasks and scoring (no search tool), so the agent must bring its own web search; scored by math equivalence (`math-verify`) of the boxed answer against the gold.\n\n## Taskset\n\n- **Source:** [`MathArena/arxivmath-training`](https://huggingface.co/datasets/MathArena/arxivmath-training)\n- **Size:** 2605 tasks\n\n## Changelog\n\n- 2026-08-31: Yield task records on demand so bounded evaluations construct only the requested prefix.\n- 2026-07-02: Initial v1 taskset.\n","encoding":"utf-8","truncated":false,"total_bytes":600},"status":null}