Sarvam AI Unveils Sarvam-1: India's 2B Parameter Model Outperforming Llama on Indic Tasks
Trained on 4 trillion tokens across 10 Indian languages, Sarvam-1 demonstrates state-of-the-art tokenization efficiency and vernacular comprehension.
Sovereign Indic Foundation Models
Bengaluru-based Sarvam AI has released Sarvam-1, a 2-billion parameter model purpose-built for Hindi, Tamil, Telugu, Kannada, Bengali, and five other Indian languages.
Tokenization Efficiency That Lowers API Bills
Global models like GPT-4o often consume 4 to 6 tokens for a single Hindi word. Sarvam-1's custom Indic tokenizer compresses regional vocabulary into 1.4 tokens per word, slashing API bills by 70%.
Written by AI Latest News Editorial Team
Dedicated team of AI researchers, benchmark evaluators, and technology journalists in Bengaluru, Hyderabad, and Delhi.
View all articles by AI Latest News Editorial Team →Discussion (0)
No comments yet. Start the conversation!
Recommended Reading
OpenAI Expands India Footprint: Lower API Pricing and Dedicated Indic Dataset Partnerships
OpenAI announces direct INR billing, local datacenter routing via Azure India Central, and collaborative academic partnerships to improve Hindi and Tamil tokenization.
Krutrim Cloud Expands Sovereign Infrastructure: Bare-Metal H100 Nodes and AI Studio
Ola's AI arm launches bare-metal GPU clusters and managed inference endpoints with native rupee billing and localized customer support.
Google DeepMind Launches Project Vaani 2: Multimodal Datasets for 50+ Indian Dialects
Collaborating with the Indian Institute of Science (IISc), Google DeepMind releases 14,000 hours of open-source conversational audio across 80 Indian districts.