A study jointly conducted by the University of Maryland and Microsoft showed that Polish has become the best language for artificial intelligence models to recognize instructions, while English only ranked sixth. The research team tested major AI language models such as OpenAI, Google Gemini, Qwen, Llama and DeepSeek. They input the same instructions in 26 different languages to evaluate the model's ability to understand and execute different languages.
The results show that Polish has an average accuracy rate of 88% in terms of command accuracy, ranking first among all languages. In comparison, English ranked only sixth with an accuracy of 83.9%, while Chinese performed worse, ranking fourth from the bottom among all tested languages.
Polish was previously considered one of the most difficult languages for humans to master. However, the study found that the AI understood and performed Polish far better than English and most other languages. This phenomenon is of particular concern because the amount of Polish training data is much lower than that of English or Chinese.
According to the test, the top ten most effective languages for AI dialogue instructions are: Polish (88%), French (87%), Italian (86%), Spanish (85%), Russian (84%), English (83.9%), Ukrainian (83.5%), Portuguese (82%), German (81%) and Dutch (80%).
The research team believes that this result is not expected by traditional intuition, and reminds technology developers and users to appropriately consider language selection when interacting with AI to improve the response efficiency and accuracy of instructions.
