Is India Keeping Up?
Tracking India's LLM release pace against the US, China, Singapore and South Korea.
Learn more278
221
99
18
6
United StatesChinaSouth KoreaSingaporeIndia
Counts are text-output LLMs released since 1 January 2026.
| Lab | Models | Release | Date | Source |
|---|---|---|---|---|
| Gnani.aiIN | 1 | Gnani Evon v3.3 30B-A3B30B hybrid Mamba-Transformer MoE (~3.5B active), continued pretraining for 11 Indian languages, open weights under Apache 2.0. | Hugging FaceGnani.ai | |
| Tech MahindraIN | 1 | Project Indus 8B8B Hindi-first education LLM built with NVIDIA NeMo under the IndiaAI Mission. | Press release | |
| BharatGenIN | 1 | Param2 17B-A2.4B Thinking17B MoE (2.4B active) trained from scratch across 22 Indian languages, open weights under a non-commercial license. | Hugging FaceBharatGen | |
| Sarvam AIIN | 2 | Sarvam 30B, Sarvam 105BTrained from scratch, open weights under Apache 2.0. | SarvamHugging Face | |
| BharatGenIN | 1 | Patram1.2 7B InstructReads Indian documents, charts and tables. Built on OLMo-7B, open weights under Apache 2.0. | Hugging Face |
Gnani.ai Tech Mahindra Project Indus 8B
BharatGen Param2 17B-A2.4B Thinking
Sarvam AI Sarvam 30B · Sarvam 105B
BharatGen Patram1.2 7B Instruct
Methodology
Counts publicly usable LLMs with text output, released by companies and labs in India, the US, China, Singapore and South Korea since 1 January 2026. Image generation, text-to-speech, speech-to-text, music generation and embedding models are not counted for now.
Why I made this