BharatGen, a government-backed AI initiative, has released Param-1, a 2.9 billion parameter bilingual large language model (LLM) built entirely from scratch. Designed as a foundational model for Indian developers and researchers, Param-1 incorporates a significant 25% Indic-language data, sharply contrasting with the 0.01% Indic content found in many global LLMs.
Branded as BharatGen Param 1 Indic Scale, the model is now available through AIKosha, the official AI repository managed by MeitY. The LLM is intended to serve as a base model that developers can fine-tune for India-specific applications, from Indic chatbots and copilots to domain-specific knowledge systems.
20 speech models across 19 Indic language variants released
Alongside Param-1, BharatGen also launched 20 new speech models designed for voice-first interfaces across 19 Indian languages. These include models under three key categories: A2TTS-v0.5 (Speaker Adaptive TTS), Speaker-Conditioned TTS (pflow), and Voicebox TTS Models. Languages covered include Hindi, Marathi, Tamil, Telugu, Bengali, Kannada, Gujarati, Malayalam, Punjabi, and more.
Also read: MeitY Shortlists Startups for IndiaAI Mission Rollout
The A2TTS models support speaker voice adaptation in nine languages, while the Speaker-Conditioned and Voicebox sets provide high-fidelity, speaker-specific voice generation. All models are available through AIKosha and have been developed using data collected directly for five Indian languages, addressing the long-standing gap in high-quality, open-access Indic speech datasets.
Public infrastructure for sovereign AI development
The Param-1 model and the speech systems are part of BharatGen’s broader strategy to establish sovereign AI infrastructure for India. Built with an investment under ₹235 crore ($27 million), the initiative is supported by a consortium of top institutions including IIT Bombay, IIT Kanpur, IIT Madras, IIT Hyderabad, IIIT Hyderabad, IIT Mandi, and IIM Indore.
BharatGen’s mission aligns with the government’s vision for self-reliant AI development and is intended to empower startups, developers, and researchers with accessible foundational tools. These models represent a step toward India’s long-term goal of developing its own AI ecosystem at scale.
