Jul 2026
RuBench: A Repository-Level Agentic Coding Benchmark with Natively Authored Russian Task Specifications
RuBench 1.0 is introduced, a benchmark of 25 tasks mined from recent fix commits in five live open-source repositories, each specified natively in Russian and judged by the upstream maintainer's regression tests, which are withheld with a SHA-256 manifest.
Evgeny Shilov
· arXiv.org · 0 citations