ResearchBenchmarksTokenization Efficiency on Punjabi (pa)Follow4.05Bytes per TokenLLaMA-43.8365.28056.7258.1695Nov 5, 2025Evaluation ResultsMethodMethodLinksBytes per TokenLLaMA-42025.114.05Gemma-32025.114.62GPT-oss2025.114.83Sarvam-2B2025.117.95Sutra2025.118.76MUTANT-Indic2025.119.4