Tokenizer Fertility and Zero-Shot Performance of Foundation Models on Ukrainian Legal Text: A Comparative Study
Tokenizer fertility varies 1.6x across foundation models on Ukrainian legal text, yet this cost-critical dimension is absent from model selection practice. Qwen 3 consumes 60% more tokens than Llama-family; NVIDIA Nemotron Super 3 (120B) outperforms Mistral Large 3 at 1/3 the cost.
Tokenizer Fertility and Zero-Shot Performance of Foundation Models on Ukrainian Legal Text
A Comparative Study
Volodymyr Ovcharov — LEX AI LLC, Kyiv, Ukraine
Abstract
Tokenizer fertility varies 1.6× across foundation models on Ukrainian legal text, yet this cost-critical dimension is absent from model selection practice. (1) Qwen 3 consumes 60% more tokens than Llama-family; (2) NVIDIA Nemotron Super 3 (120B) outperforms Mistral Large 3 (5.6× more parameters) at 1/3 the cost; (3) few-shot degrades by up to 26pp on Ukrainian.