News

Source — Source: TH ///Category — GS Paper 2 ///Published — 2026-02-19 ///Exam — Both Prelims and Mains ///Tags — Rapid Fire CA, Quick Facts For Prelims, GS Paper - 2 ///Source — Source: TH ///Category — GS Paper 2 ///Published — 2026-02-19 ///Exam — Both Prelims and Mains ///Tags — Rapid Fire CA, Quick Facts For Prelims, GS Paper - 2 ///

Source: TH · GS Paper 2

Sarvam-30B and Sarvam-105B

Bengaluru-based Sarvam AI unveiled two large language models, Sarvam-30B and Sarvam-105B, aimed at mastering all 22 Indian languages. The initiative, backed by the IndiaAI Mission, underscores India's push for a sovereign, open‑source AI ecosystem to enhance governance, digital inclusion, and strategic autonomy.

2026-02-19

Published

Source: TH

Source

Quiz Questions

Article

Key Facts and Data Points

Models: Sarvam-30B (30 billion parameters) and Sarvam-105B (105 billion parameters)
Token context windows: 32,000 tokens (30B) and 128,000 tokens (105B)
Language coverage: All 22 Indian languages with voice‑first optimisation
Architecture: Mixture‑of‑Experts (MoE) – activates only relevant experts during inference
Open‑source: Code and weights released publicly
Training infrastructure: GPUs via IndiaAI Mission’s common compute programme
Related initiatives: India’s first sovereign LLM ecosystem (120 B‑parameter model) for governance; other sector‑specific models (Soket, Gnani, Gan AI)

Background and Context

Unveiled at India‑AI Impact Summit 2026 in Bengaluru.
Introduced Vikram, a multilingual chatbot named after physicist Vikram Sarabhai, symbolising indigenous scientific innovation.
Launched as global players (e.g., OpenAI) introduce benchmarks like IndQA focusing on Indian language understanding.
Addresses the “data scarcity” problem in Indic languages, reducing reliance on foreign AI models.

Significance for India / Governance / Policy

Sovereign AI: Aligns with the Government’s vision of an indigenous AI ecosystem, reducing strategic dependence on external providers.
Public‑private partnership: Demonstrates effective utilization of the IndiaAI Mission’s shared compute resources.
Governance applications: Planned 120 B‑parameter open‑source model for public services, enhancing transparency, accessibility, and efficiency.
Economic impact: Open‑source models can catalyse local startups, research, and job creation in the AI sector.
Social inclusion: Multilingual capability promotes digital inclusion across diverse linguistic groups, supporting the Digital India agenda.

Related Constitutional / Legal Provisions

Information Technology Act, 2000 – governs electronic data, cybersecurity, and digital transactions; relevant for AI deployment and liability.
Data Protection Bill (pending) – will regulate personal data handling in AI applications, emphasizing consent and privacy.
National AI Strategy (NITI Aayog) – emphasizes indigenous development, ethical AI, capacity building, and sector‑specific use‑cases.