Sarvam open-sources 30B, 105B reasoning models; here’s what it means

Indian AI startup Sarvam has open-sourced its 30B and 105B reasoning models, making them accessible for research and development, with a focus on efficiency and Indian language capabilities.

• •Sarvam has open-sourced its 30B and 105B reasoning models, named Sarvam 30B and Sarvam 105B.
• •These models are trained from scratch in-house and excel in global benchmarks and Indian languages.
• •The 30B model uses a Mixture-of-Experts (MoE) architecture for efficient reasoning, activating about 1 billion parameters per token.