Ukrainian AI experts have launched the first open leaderboard for evaluating Ukrainian handwritten text recognition (OCR/HTR) models — RUKOPYS Leaderboard — on the Hugging Face platform. The initiative grew out of the spring Handwritten to Data competition on Kaggle and aims to create a transparent standard for testing such systems.
Ukrainian AI experts have launched the first open leaderboard for evaluating Ukrainian handwritten text recognition (OCR/HTR) models — RUKOPYS Leaderboard — on the Hugging Face platform. The initiative grew out of the spring Handwritten to Data competition on Kaggle and aims to create a transparent standard for testing such systems.
The launch of the platform was announced by AI/ML Lead of the educational project "Mriya" Dmytro Voytech on his LinkedIn.
The test standard was created on the basis of RUKOPYS — the first large open dataset of Ukrainian handwritten text, prepared jointly with the Ukrainian Catholic University (UCU). The sample included 385 test pages (8290 text regions) with documents from the period from 1919 to 2025: archival records, sheets of the national unity radio dictation, university exams, and school notebooks.
The benchmark has already tested 43 models and systems in 94 configurations. Among them are commercial models via API, open-source solutions, and specialized systems from the competition participants. The evaluation takes into account the completeness of detection, classification accuracy, character error rate, and the estimated cost of processing 1,000 pages.
The leader of the RUKOPYS Leaderboard by composite score was Google's Gemini 4 Argon model with a score of 0.9387 and a processing cost of $53.08 per thousand pages. It is followed by other versions of Gemini, as well as solutions from Anthropic and OpenAI.
According to Dmytro Voytech, the test results proved that high recognition quality does not necessarily require high costs. Some commercial models at the cheapest settings are almost as good as the most expensive configurations, and compact specialized systems from participating developers compete with large general-purpose models.
All project materials (raw model responses for each page, evaluation code, prompts, and precise launch settings) have been made publicly available on Hugging Face for independent review. The developers also invite the AI community to contribute their own models for comparison.
The best of the tested models are already being used in practice to digitize Ukrainian state archives. In parallel, the team is working on a new generation of models that will combine the best solutions of the competition participants and a much larger amount of training data.
The project was implemented with the support of UCU (dataset hosting and legal support), the AI HOUSE community (organization of the competition on Kaggle), and the Center of Excellence in AI at the Ministry of Digital Transformation, headed by Roman Kysly.
