Robots Atlas>ROBOTS ATLAS
Evaluation

Minerva Math

2022ActiveUpdated: 22 September 2026Published
Key innovation
Bundling quantitative STEM reasoning into a coherent evaluation suite (the Minerva model).
Category
Evaluation
Abstraction level
Building block
Operation level
Evaluation (runtime)
Use cases
Quantitative reasoning evaluationBenchmarking math modelsSTEM problems

How it works

The model solves MATH/GSM8K/STEM problems with step-by-step reasoning; the score is final-answer accuracy.

Problem solved

There was no consistent reference point for quantitative, multi-step STEM reasoning.