ResearchBenchmarksTokenization Efficiency on Sanskrit sanFollow5.45Bytes per TokenLLaMA-45.326.19757.0757.9525Nov 5, 2025Evaluation ResultsMethodMethodLinksBytes per TokenLLaMA-42025.115.45GPT-oss2025.115.79Sarvam-2B2025.116.03Sutra2025.116.04Gemma-32025.116.66MUTANT-Indic2025.118.7