ResearchBenchmarksTokenization Efficiency on Assamese (as)Follow3.84Bytes-per-TokenLLaMA-43.62885.05446.487.9056Nov 5, 2025Evaluation ResultsMethodMethodLinksBytes-per-TokenLLaMA-42025.113.84Sarvam-2B2025.113.92GPT-oss2025.116.36Gemma-32025.116.37Sutra2025.118.04MUTANT-Indic2025.119.12