Processing Uzbek texts using artificial intelligence costs 1.55–2.17 times more than in English. Analysts from Gorgona AI reached this conclusion in their study “One Meaning, Different Price. How Much Does the Uzbek Language Cost for AI?”. The experts compared five major providers: OpenAI, Anthropic, Google, Qwen, and DeepSeek. For the analysis, they used 1,337 parallel texts with identical content in different languages, testing them across 15 tokenizers and real model responses.

The reason lies in the specifics of tokenization—splitting text into fragments processed by language models. Because training data contains significantly more English, common English words are often encoded as a single token. Uzbek words with numerous affixes are split into nearly three tokens on average. Since providers charge per token regardless of the language, request processing costs increase. In addition, less information fits into the context window, and response generation takes more time.

The difference depends on the model and the type of text. Materials originally written in Uzbek usually end up closer to the lower end of the range, while translations from English are closer to the upper end. On the FLORES corpus, which includes news, encyclopedic articles, and travel guides, the difference ranged from 1.78 times for OpenAI to 2.13 times for DeepSeek. Russian is also more expensive than English—by 1.35–1.60 times. Meanwhile, Uzbek is 1.24–1.54 times more expensive than Russian. The influence of genre turned out to be minor: the ratio varies by no more than 8%.