ResearchBenchmarksTokenization Efficiency on Kannada knFollow5.33Bytes per TokenLLaMA-45.13126.47317.8159.1569Nov 5, 2025Evaluation ResultsMethodMethodLinksBytes per TokenLLaMA-42025.115.33Gemma-32025.116.73GPT-oss2025.116.74Sutra2025.118.38Sarvam-2B2025.119.07MUTANT-Indic2025.1110.3