NZ121/3 (15.1/20 ov)
IND
NAM49/2 (10.5/50 ov)
IRE
MOZ
UGA
AVL
MNC
RAY
ELC
GMB
ZMB
NGA
LES
CAG
BOL
COMO
SAS
FCA
FCU
Ipswich1
Nottm Forest1
Alavés1
Málaga1
Torino3
Monza2
Brest1
Nice2
WA
NSW199/3 (31/50 ov)
DEN
KUW
KENYA
QAT
SWD-W
EPR-W
TTN-W
WPR-W
NZ-A135/3 (41 ov)
IND-A113
DHA
RAJ
MYMS
CHATT
RAN
BAR
SYL
KHU
MEX-W
ARG-W
UAE
NED
PAK
ENG
IND-W
ZIM-W
AFG170/3 (34.3/50 ov, target 170)
ZIM169
TAS137 (25.2/35 ov, target 276)
VIC304/7 (50 ov)
CHE0
DYI0
CHO1
QIN0
HEN1
BEI0
LIA1
ZHE0
QWC1
YUN0
SHX0
Shanghai Shenh0
TIG0
SHP0
WTT1
SHT3
KOE
S04
M05
SVW
RBL
ELV
Stuttgart2
Gladbach3
Al Ahli4
Al Khaleej1
India’s homegrown AI ambitions just reached a new milestone. At the India AI Impact Summit 2026 in New Delhi, BharatGen — the government-backed sovereign AI initiative under IIT Bombay’s Technology Innovation Hub — unveiled Param2 17B MoE: a 17-billion-parameter Mixture-of-Experts multilingual foundational AI model built entirely for India, by India.
BharatGen Param2 17B MoE: Key Details
This is not just another large language model. Param2 17B MoE is a sovereign AI model — trained, owned, and open-sourced within India — purpose-built for the country’s linguistic diversity and critical sector needs.
| Detail | Info |
|---|---|
| Model Name | BharatGen Param2 17B MoE |
| Parameters | 17 billion (Mixture-of-Experts architecture) |
| Backed by | Department of Science & Technology (DST), Govt. of India |
| Built at | IIT Bombay Technology Innovation Hub |
| NVIDIA Partnership | Built on NVIDIA AI Enterprise, NeMo, NeMo-RL, Base Command Manager |
| Target Languages | Multiple Indic languages |
| Key Sectors | Governance, education, healthcare, agriculture, enterprise |
| Open Source | Yes — available on Hugging Face |
| Announced at | India AI Impact Summit 2026, Bharat Mandapam, New Delhi |
The Mixture-of-Experts (MoE) architecture is significant — it allows the model to activate only relevant “expert” sub-networks for each task, making it far more computationally efficient than a dense 17B model while maintaining high performance across diverse language and domain tasks.
Why This Matters for India’s AI Future
BharatGen’s mandate goes beyond building a model — it’s about AI sovereignty. By open-sourcing Param2 17B MoE on Hugging Face alongside full documentation and post-training workflows, BharatGen is enabling India’s developer ecosystem to immediately build, fine-tune, and deploy India-centric AI applications — without dependency on foreign AI infrastructure or proprietary platforms.
The NVIDIA partnership provides the backbone: training pipelines powered by NVIDIA NeMo open libraries and Base Command Manager ensure the model is production-grade and scalable from day one.
With use cases spanning government services, multilingual education tools, rural healthcare, and agriculture advisory systems, Param2 17B MoE is built to reach citizens who have historically been underserved by English-first AI systems.
According to BharatGen’s official Hugging Face repository, the model and documentation are openly available for developers right now.
Check out our India AI and sovereign tech coverage on TechnoSports and our AI model news for more.
