OpenModelsHQ

India

Open models from India

We track 21 open-weight models from 2 labs in India. They account for 3 of 205 tracked releases (1%) from Dec 2025 to Nov 2026. The newest is bhili-tts, listed on 21 May 2026.

Explore the models
Bengaluru skyline
Bengaluru skyline Photo credit

Latest models

Newest listed first
  1. bhili-ttsAI4Bharat · 938M
  2. sarvam-105bSarvam AI · 106B
  3. sarvam-30bSarvam AI · 32.2B
  4. sarvam-mSarvam AI · 23.6B
  5. indic-conformer-600m-multilingualAI4Bharat · Size not listed
  6. IndicF5AI4Bharat · 351M
More models from India
  1. indic-seamlessAI4Bharat · 1.5B
  2. vits_rasa_13AI4Bharat · 40M
  3. indic-parler-tts-pretrainedAI4Bharat · 938M
  4. indic-parler-ttsAI4Bharat · 938M
  5. sarvam-1Sarvam AI · 2.5B
  6. indicconformer_stt_ur_hybrid_ctc_rnnt_largeAI4Bharat · Size not listed

Cheapest tracked APIs

No hosted API prices recorded. The lab pages link to the available weights.

How the work developed

AI4Bharat released IndicConformer for speech recognition across India’s scheduled languages. The multilingual checkpoint uses a Conformer architecture and supports transcription with alternative decoding methods. Its model card identifies an MIT licence and gives instructions for running the released weights.

Sarvam’s Sarvam-1 work focused on a language model trained for Indian languages alongside English. Sarvam documented training on Yotta’s Shakti infrastructure using NVIDIA NeMo, and published the weights and tokenizer. Together, these releases cover spoken-language recognition and text generation. Their model cards describe the languages and training choices directly, so readers can check support for the language they intend to use.

Sources and counting method

Counts cover selected repositories, including model variants. Dates come from Hugging Face repository creation dates and may differ from public announcements. Location follows the lab record, not the nationality of individual contributors.

Sources