ResearchBenchmarksTokenization Efficiency on Programming CodeFollow1.95Bytes per TokenSarvam-2B1.9142.1572.42.643Nov 5, 2025Evaluation ResultsMethodMethodLinksBytes per TokenSarvam-2B2025.111.95Sutra2025.111.97Gemma-32025.112.33GPT-oss2025.112.77MUTANT-Indic2025.112.84LLaMA-42025.112.85