Benchmarking large language models on the Ukrainian Krok 2 licensing examination in physical therapy: accuracy, consistency, and agreement
- 1Uzhhorod National University
Abstract
AimTo evaluate the accuracy, run-to-run consistency, and agreement of large language models on the Ukrainian Integrated Licensing Examination Krok 2 for the Specialty of Physical Therapy.
Materials and MethodsFive cloud-based LLMs (ChatGPT-5.5, Claude 5 Sonnet, Gemini 3.5 Flash, Grok 4, DeepSeek-V3) were evaluated using 150 Ukrainian-language multiple-choice questions from the official Krok 2 database. Each model completed two independent runs with identical inputs. Accuracy (with 95% confidence intervals, Wilson method), run-to-run consistency, and agreement were assessed. Statistical analysis included Cohens {kappa}, Cochrans Q and pairwise McNemar tests with Holm-Bonferroni correction.
ResultsAll models achieved >80% accuracy in both runs, exceeding the passing threshold of [≥]64%. In the first run, accuracy ranged from 80.00% (DeepSeek-V3) to 93.33% (Gemini 3.5 Flash); in the second run, from 82.67% (Claude 5 Sonnet) to 88.67% (Gemini 3.5 Flash). In the first run significant differences were observed between Gemini 3.5 Flash and Grok 4 (93.33% vs 84.00%; p=0.0022) and DeepSeek-V3 (93.33% vs 80.00%; p=0.0001). There were no statistically significant differences between models in the second run. R2R consistency exceeded 90% for all models with the highest in ChatGPT-5.5 (98.67%) and lowest in DeepSeek-V3 (90.00%). Cohens {kappa} ranged from 0.64 to 0.94, indicating substantial to almost perfect agreement. The stability of LLMs response correctness across two runs ranged from 77.33% (DeepSeek-V3) to 88.00% (Gemini 3.5 Flash).
ConclusionsAll LLMs demonstrated high performance on Ukrainian-language Krok 2 examination tasks. Gemini 3.5 Flash showed the highest accuracy in both runs and response correctness stability while ChatGPT-5.5 exhibited the greatest response consistency and agreement. Although the accuracy and stability of LLM responses may be interrelated, they represent distinct aspects of model performance. This highlights the need for a comprehensive approach to evaluating LLMs. The findings indicate the potential of LLMs as an auxiliary tool in the professional practice of physical therapy specialists.
Related
- Research Preparedness Among Osteopathic Medical Students in the Post-Pass/Fail Licensing Examination Era: Bridging the Gap Between Rising Expectations and Student Confidence
- Comparative Evaluation of a System One Model and a General-Purpose Large Language Model on the Korean Physical Therapist Licensing Examination
- The lived experiences of diploma nurses who did not pass their professional licensing examination on the first attempt: A qualitative phenomenological study
- Midlife mobility assessment from the pocket: agreement and consistency between smartphone placement configurations
- mTORC1-dependent control of Hedgehog signaling mediates nutrient-responsive licensing of intestinal stem cell differentiation
BibTeX
@misc{filak2026benchmarking,
title = {{Benchmarking large language models on the Ukrainian Krok 2 licensing examination in physical therapy: accuracy, consistency, and agreement}},
author = {Y. F. Filak and Y. O. Mykhalko and M. V. Sabadosh},
howpublished = {medRxiv},
year = {2026},
month = sep,
doi = {10.64898/2026.09.21.26363368},
url = {https://www.medrxiv.org/content/10.64898/2026.09.21.26363368v1},
note = {CC BY-NC-ND},
}
RIS
TY - UNPB
TI - Benchmarking large language models on the Ukrainian Krok 2 licensing examination in physical therapy: accuracy, consistency, and agreement
AU - Y. F. Filak
AU - Y. O. Mykhalko
AU - M. V. Sabadosh
PY - 2026
DA - 2026/09/21
PB - medRxiv
DO - 10.64898/2026.09.21.26363368
UR - https://www.medrxiv.org/content/10.64898/2026.09.21.26363368v1
AB - AimTo evaluate the accuracy, run-to-run consistency, and agreement of large language models on the Ukrainian Integrated Licensing Examination Krok 2 for the Specialty of Physical Therapy. Materials and MethodsFive cloud-based LLMs (ChatGPT-5.5, Claude 5 Sonnet, Gemini 3.5 Flash, Grok 4, DeepSeek-V3) were evaluated using 150 Ukrainian-language multiple-choice questions from the official Krok 2 database. Each model completed two independent runs with identical inputs. Accuracy (with 95% confidence intervals, Wilson method), run-to-run consistency, and agreement were assessed. Statistical analysis included Cohens {kappa}, Cochrans Q and pairwise McNemar tests with Holm-Bonferroni correction. ResultsAll models achieved >80% accuracy in both runs, exceeding the passing threshold of [≥]64%. In the first run, accuracy ranged from 80.00% (DeepSeek-V3) to 93.33% (Gemini 3.5 Flash); in the second run, from 82.67% (Claude 5 Sonnet) to 88.67% (Gemini 3.5 Flash). In the first run significant differences were observed between Gemini 3.5 Flash and Grok 4 (93.33% vs 84.00%; p=0.0022) and DeepSeek-V3 (93.33% vs 80.00%; p=0.0001). There were no statistically significant differences between models in the second run. R2R consistency exceeded 90% for all models with the highest in ChatGPT-5.5 (98.67%) and lowest in DeepSeek-V3 (90.00%). Cohens {kappa} ranged from 0.64 to 0.94, indicating substantial to almost perfect agreement. The stability of LLMs response correctness across two runs ranged from 77.33% (DeepSeek-V3) to 88.00% (Gemini 3.5 Flash). ConclusionsAll LLMs demonstrated high performance on Ukrainian-language Krok 2 examination tasks. Gemini 3.5 Flash showed the highest accuracy in both runs and response correctness stability while ChatGPT-5.5 exhibited the greatest response consistency and agreement. Although the accuracy and stability of LLM responses may be interrelated, they represent distinct aspects of model performance. This highlights the need for a comprehensive approach to evaluating LLMs. The findings indicate the potential of LLMs as an auxiliary tool in the professional practice of physical therapy specialists.
ER -