List your company

Per-character tokenization on GSM8K (train)

0.381Tokens per Character Ratio (GSM8K train)Sarvam-1
0.29780.31940.3410.3626May 28, 2026

Evaluation results

MethodLinks
Sarvam-1Vocab=68K2026.050.381
Tekken/Sarvam-mVocab=131K2026.050.351
Sarvam-30BVocab=262K2026.050.351
BrahmicTokenizer-131KVocab=131K2026.050.301