RuBench: leaderboard
RuBench 1.0 (2026) measures deployed coding-agent configurations (agent CLI, model and reasoning effort) on 25 repository-level bug-fix tasks in Python, PHP, TypeScript and JavaScript whose specifications are written natively in Russian, graded by withheld maintainer regression tests. Pass@1 over three runs in percent, higher is better.
Metric: Pass@1 (%; all 25 tasks; pass@1, mean share of tasks resolved over three independent runs, judged by the withheld upstream maintainer regression tests; 25 repository-level fix tasks from aiohttp, aiogram, Laravel, NestJS and Fastify with task statements natively written in Russian; deployed agent CLI, model and reasoning effort as one configuration, product defaults otherwise, network access allowed). Source: arxiv.org. Saturation forecast: Around February 2027. 6 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.8 (xHigh) | 78.7 | #56 (Claude Opus 4.8) |
| 2 | Claude Sonnet 5 (xHigh) | 74.7 | #111 (Claude Sonnet 5) |
| 3 | GPT-5.5 (xHigh) | 66.7 | #49 (GPT-5.5) |
| 4 | Gemini 3.1 Pro (Preview) (High) | 49.3 | #94 (Gemini 3.1 Pro (Preview)) |
| 5 | Gemini 3.5 Flash (High) | 46.7 | #101 (Gemini 3.5 Flash) |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=rubench · How It Works · Data refreshed daily, snapshot 2026-11-13.