Benchmark
BudHist Benchmark
How much Buddhist history do language models know across regions and historical periods?
Explore the original evidence
Explore the original evaluation, its methods, and reported findings.
Open evalCoverage
The October 2025 test had 150 questions, April 2026 had 210, and August 2026 had 250. The August test added 40 questions on Japanese Buddhist history.
Methodology
Each question offers four answers. The project compares local and hosted open-weight models and includes closed models for reference. Question coverage changes between tests. Compare models within the same dated set.