# SAIR Benchmark

> SAIR's public benchmarks for AI on mathematical reasoning. Hosted by SAIR (the Foundation for Science and AI Research).

The current release is the Equational Theories Benchmark — measuring how well leading AI models decide whether one equation implies another, across problem sets at multiple difficulty levels and comparable reasoning/temperature settings. Models are accessed via OpenRouter; problem sets, runs, leaderboards, and the evaluation prompt are published on Hugging Face.

## More from SAIR

- [Competitions](https://competition.sair.foundation): Open mathematics and AI challenges.
- [SAIR Foundation](https://sair.foundation): Mission, programs, and events.
- [Hugging Face](https://huggingface.co/SAIRfoundation): Datasets, models, and benchmark data.
- [Blog](https://blog.sair.foundation/)
- Contact: press@sair.foundation