BharatGen
Government-funded Indic language and speech models led by IIT Bombay
Overview
BharatGen is an IIT Bombay-led, government-funded consortium building Indian-language AI: the Param2 17B mixture-of-experts LLM covering 22 scheduled languages, Shrutam-2 speech recognition and Sooktam-2 text-to-speech. Weights are free on Hugging Face, but the flagship models are licensed for non-commercial research only.
Pricing
Pricing shown for reference only. These figures reflect RECATOOLS research as of 12 Oct 2026 and may be out of date or incomplete. This is not financial or purchasing advice — always confirm the current price on the provider’s official website before making any decision.
Use cases
What you can produce with BharatGen
- Param2 17B mixture-of-experts LLM covering English and 22 scheduled Indian languages
- Shrutam-2 speech recognition and Sooktam-2 text-to-speech in 12 languages each
- Param-1 models and AgriParam, FinanceParam, LegalParam and AyurParam domain models
- Patram document-understanding model
- BhashaBench benchmarks for agriculture, law, finance and Ayurveda
- Weights on Hugging Face, with the flagship models under a non-commercial licence
BharatGen is a nine-institution research consortium led from IIT Bombay with Indian government funding. Its flagship Param2 is a 17B mixture-of-experts model trained from scratch and covering English plus all 22 scheduled Indian languages, released with Indic speech models and domain benchmarks.
The catch is the licence: Param2 and the newest speech models are non-commercial research only and need written approval for commercial or public deployment; just the older Param-1 2.9B, Patram and the domain models are Apache 2.0. A strong research resource, not something a business can deploy freely.
What we found
BharatGen is a government-funded research programme rather than a software vendor. India's Department of Science and Technology approved it in July 2024 and it launched that October, run by a nine-institution consortium led from IIT Bombay and operated through the BharatGen Technology Foundation on the IIT Bombay campus in Powai. Funding came in two pieces: ₹235 crore from DST's cyber-physical systems mission, as reported in the press, then ₹988.6 crore awarded under the IndiaAI Mission in September 2025. Earlier versions of this page gave a combined figure we could not trace.
The headline model is Param2, a 17-billion-parameter mixture-of-experts model that activates 2.4 billion parameters per token, was trained from scratch on about 22 trillion tokens and covers English plus all 22 scheduled Indian languages. It was unveiled at the India AI Impact Summit in February 2026 alongside Shrutam-2 speech recognition and Sooktam-2 text-to-speech, each covering 12 languages, and the Patram document model. Older releases include the Param-1 family and domain models for agriculture, finance, law and Ayurveda.
The licensing matters more than the press coverage suggests. Param2, Shrutam-2, Sooktam-2 and the newer Param-1 checkpoints ship under the BharatGen Research License, which allows non-commercial research only and requires written approval before commercial use or public deployment. Only the earlier Param-1 2.9B Instruct, Patram and the four domain models are Apache 2.0. The Param2 card also calls it an early post-training checkpoint needing further alignment before production.
BharatGen's own benchmark table is candid: Param2 scores 57.8 on English MMLU, below gpt-oss-20b and Qwen1.5-MoE, but 59.2 on Hindi MMLU, ahead of every model it lists. Its BhashaBench suite, built from Indian exams in agriculture, law, finance and Ayurveda, was used in a paper testing more than 29 models. For Indian-language research the releases are substantial; businesses should read the licence before building anything.
Compiled by RECATOOLS from public sources: vendor documentation and pricing, code repositories and published reporting.
Frequently asked questions
About this listing
This entry was compiled from publicly available data including BharatGen's official website, press releases, documentation, and reputable third-party publications. RECATOOLS is not affiliated with BharatGen unless explicitly stated.
Third-party AI tools update their pricing, features, availability, and policies frequently. Information here may be outdated by the time you read this — we make reasonable efforts to keep listings current, but cannot guarantee absolute accuracy.
For the latest details, please refer to BharatGen directly →
Spotted something out of date? Suggest an update →
Related reading
Alternatives to BharatGen
More in LLMs & Chat
