Benchmark RadarMIRA-Math
MIRA-Math evaluates mathematical reasoning where each problem is missing exactly one necessary atomic fact that must be requested in natural language under a strict budget, then integrated into an exact answer. It contains 2,310 instances across 22 typed mathematical families.
- Released
- 2026-07-08
- Readiness
- Paper only
- Primary field
- General AI
Why it matters
Isolates the diagnostic capability of minimal information requesting separate from broader tool use or long-horizon dialogue, revealing that request success and final-answer accuracy are separable.
Motivation
Mathematical reasoning benchmarks typically provide all facts needed to solve each problem, while interactive benchmarks often mix reasoning with tools, retrieval, and long-horizon dialogue.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.