Atomic Canyon, together with Idaho National Laboratory (INL), is launching NuclearBench: a public benchmark for how well AI models answer technical questions from real nuclear regulatory and engineering documents.
Why a nuclear-specific benchmark
General AI benchmarks don’t tell you whether a model can be trusted with dense nuclear documentation. Real documents combine scan artifacts, buried cross-references, specialized vocabulary, and revision-sensitive limits. NuclearBench tests whether models can connect those scattered facts while staying grounded in the source.
What’s on the site
- A public leaderboard comparing model performance
- The dataset and what it covers
- How answers are scored, with substance and faithfulness assessed separately
- Methodology for running the benchmark yourself
NuclearBench grew out of our benchmark program with INL, and is designed to benefit the broader nuclear community.
Explore the results at nuclearbench.com.