ResearchBenchmarksTokenization Efficiency on Hindi hiFollow7.37Bytes per TokenLLaMA-47.22448.20729.1910.1728Nov 5, 2025Evaluation ResultsMethodMethodLinksBytes per TokenLLaMA-42025.117.37GPT-oss2025.117.83Sutra2025.118.33Sarvam-2B2025.118.92Gemma-32025.119.16MUTANT-Indic2025.1111.01