ResearchBenchmarksTokenization Efficiency on Marathi mrFollow6.49Bytes per TokenLLaMA-46.30527.55268.810.0474Nov 5, 2025Evaluation ResultsMethodMethodLinksBytes per TokenLLaMA-42025.116.49GPT-oss2025.116.94Sutra2025.118.35Gemma-32025.118.9Sarvam-2B2025.119.46MUTANT-Indic2025.1111.11