LegalBench-RU
Testing AI models on legal tasks
PythonAugust 2026LegalBench-RUоценка моделейправо
About the project
LegalBench-RU is a Russian-language legal task set with tools for comparing models under the same conditions. The corpus has a fixed version and contents, a deterministic scorer evaluates answers, and a dashboard shows coverage, errors and results by task category.
Incomplete or incomparable runs are excluded from the ranking. Published scores remain preliminary until lawyers have reviewed the tasks and answers.
Features
- Versioned corpus and verifiable release manifest
- Shared run protocol and deterministic scoring
- Dashboard with coverage and category breakdowns