📊 Open-weight releases
| Rank | Model | Primary focus | Link | downloads* | Maintainer(s) |
|---|---|---|---|---|---|
| 1 | IndicTrans2 (family)** | Multilingual translation stack covering 22 Indic languages | https://huggingface.co/AI4Bharat/indictrans2-en-indic-1B | 110,328 | AI4Bharat |
| 2 | sarvamai/sarvam-translate | Gemma-3 based multimodal translation & assistant for Indic workflows | https://huggingface.co/sarvamai/sarvam-translate | 25,127 | Sarvam AI |
| 3 | sarvamai/sarvam-m | 24B chat-tuned LLM spanning 11 Indian languages | https://huggingface.co/sarvamai/sarvam-m | 2,790 | Sarvam AI |
| 4 | sarvamai/sarvam-1 | Llama-2 derived multilingual chat model (bn/en/gu/hi/kn/ml/mr/or/pa/ta/te) | https://huggingface.co/sarvamai/sarvam-1 | 2,504 | Sarvam AI |
| 5 | sarvamai/OpenHathi-7B-Hi-v0.1-Base | Hindi-first 7B base model for Indic workloads | https://huggingface.co/sarvamai/OpenHathi-7B-Hi-v0.1-Base | 1,664 | Sarvam AI • AI4Bharat • Bhashini |
| 6 | sarvamai/sarvam-0.5 | 2B instruction-tuned model optimized for code-mixed Indic chat | https://huggingface.co/sarvamai/sarvam-0.5 | 1,293 | Sarvam AI |
| 7 | aashay96/indic-gpt | GPT-2 scale baseline for Indic text generation experiments | https://huggingface.co/aashay96/indic-gpt | 175 | Community (aashay96) |
ai4bharat/indictrans2-en-indic-1B (55,506), ai4bharat/indictrans2-indic-en-1B (17,999), ai4bharat/indictrans2-indic-indic-1B (13,828), ai4bharat/indictrans2-indic-indic-dist-320M (8,304), ai4bharat/indictrans2-indic-en-dist-200M (7,623), ai4bharat/indictrans2-en-indic-dist-200M (7,068).
🚧 Closed / sovereign programs without public checkpoints
- Hanooman — SML India, Seetha.ai, IIT Bombay: Multilingual suite (22 Indic + 10 global languages); enterprise deployments, no Hugging Face release.
- Bhashini / AIRAWAT LLM — MeitY, C-DAC: National language mission building sovereign inference infrastructure; models gated within the NLTM stack.
- Krutrim — Ola / Krutrim SI Designs: Full-stack AI platform with proprietary pretraining on Indian corpora; checkpoints not open-sourced.
- Dhruva — NVIDIA, C-DAC, MeitY: AIRAWAT-backed compute cluster enabling large-scale sovereign model training rather than a single public LLM.
- Kosmos India (AIRAWAT) — NIC, C-DAC Pune: Multimodal research initiative (text + vision) currently limited to internal partners.
- BharatGPT — Reliance Jio, IIT Bombay: In-development enterprise LLM suite; benchmarks and weights remain undisclosed.
🧠Govt & Research Infrastructure Enablers
- Bhashini Mission: Multilingual datasets, speech and translation APIs — https://bhashini.gov.in
- AI4Bharat: Open research initiative for Indic NLP and LLMs — https://ai4bharat.iitm.ac.in/