India

AI models, vendors, and hardware buying realities from India. Curated by myaihardware.com.

Models

Vendors

Languages

Why India matters for AI builders

India is the world's largest market for AI inference at the edge, yet builders face a hard reality: NVIDIA's A100 and H100 GPUs remain severely restricted and expensive here, forcing reliance on cloud rentals or older chips like the A2. The dominant model families are fine-tuned variants of Llama, Gemma, and Qwen — open-weight, not proprietary — so hardware must optimize for batch inference and quantization rather than massive pre-training. The underrated angle is India's uniquely fragmented telecom and power infrastructure, which demands AI models that run reliably on intermittent 4G and low-wattage CPUs, not just data-center V100s.

Market context

India's AI ecosystem is being propelled by the IndiaAI Mission's ₹10,372 crore funding, which is driving government-led compute infrastructure and application development. The vast scale of India Stack, Aadhaar, and DigiLocker, combined with 22 official languages, creates a massive demand for multilingual models that can serve diverse populations. Policies in 2025 reducing GPU import duties, alongside major compute build-outs by Yotta and Reliance Jio, are lowering barriers for domestic AI training and inference. Homegrown players like Sarvam, Krutrim, and AI4Bharat are forming a reliable ecosystem with strong Indic-language coverage, positioning India as a key hub for localized AI solutions.

Models (20 of 20)

Filter by vendor, license, or parameter range.

LicenseVendorParameter range

Sarvam-1

Sarvam AI · India

Open

Params: 1B
Context: 4,096
Min VRAM: 2 GB
GGUF: Yes

Strong multilingual Indic language support with efficient 1B parameter size.

Sarvam-2B

Sarvam AI · India

Open

Params: 2B
Context: 8,192
Min VRAM: 4 GB
GGUF: Yes

Improved context length and parameter count for better Indic language generation.

Krutrim-1

Krutrim AI (Ola) · India

Open

Params: 7B
Context: 4,096
Min VRAM: 14 GB
GGUF: Yes

Broad Indic language coverage with 10 languages and strong generation.

Krutrim-2

Krutrim AI (Ola) · India

Open

Params: 13B
Context: 8,192
Min VRAM: 26 GB
GGUF: Yes

Larger model with improved context and performance across Indic languages.

OpenHathi-7B

AI4Bharat · India

Open

Params: 7B
Context: 4,096
Min VRAM: 14 GB
GGUF: Yes

Strong Hindi-English bilingual performance with open-source availability.

IndicTrans2

AI4Bharat · India

Open

Params: 1B
Context: 512
Min VRAM: 2 GB
GGUF: No

Supports 22 Indic languages for translation with high accuracy.

IndicBART

AI4Bharat · India

Open

Params: 400M
Context: 512
Min VRAM: 1 GB
GGUF: No

Efficient multilingual model for Indic language generation and summarization.

IndicBERT

AI4Bharat · India

Open

Params: 100M
Context: 512
Min VRAM: 0.5 GB
GGUF: No

Lightweight multilingual encoder for Indic language understanding tasks.

Hanooman

TII (Technology Innovation Institute) · India

Open

Params: 7B
Context: 4,096
Min VRAM: 14 GB
GGUF: Yes

Multilingual Indic support with strong generation and open-source.

BharatGPT

CoRover.ai · India

Restricted

Params: 7B
Context: 4,096
Min VRAM: 14 GB
GGUF: No

Enterprise-focused with broad Indic language support and conversational AI.

Param-1

ParamAI · India

Open

Params: 1B
Context: 4,096
Min VRAM: 2 GB
GGUF: Yes

Efficient small model for Indic languages with open-source availability.

SUTRA

Tata Consultancy Services (TCS) · India

Restricted

Params: 7B
Context: 8,192
Min VRAM: 14 GB
GGUF: No

Enterprise-grade with long context and broad Indic language support.

Nallah

Adya AI · India

Open

Params: 1B
Context: 4,096
Min VRAM: 2 GB
GGUF: Yes

Lightweight Indic model with efficient performance and open-source.

BharatLLM

IIT Bombay · India

Open

Params: 7B
Context: 4,096
Min VRAM: 14 GB
GGUF: Yes

Academic research model with broad Indic language support.

AI4Bharat (IIT Madras)

IIT Madras · India

Open

Params: 7B
Context: 4,096
Min VRAM: 14 GB
GGUF: Yes

Research-driven model with strong Indic language capabilities.

Llama-3-Tamil

Abhinand · India

Open

Params: 8B
Context: 8,192
Min VRAM: 16 GB
GGUF: Yes

Fine-tuned Llama 3 for Tamil with strong bilingual performance.

Telugu-Llama3

Community · India

Open

Params: 8B
Context: 8,192
Min VRAM: 16 GB
GGUF: Yes

Fine-tuned Llama 3 for Telugu with strong bilingual capabilities.

Hi-Llama-7B

Community · India

Open

Params: 7B
Context: 4,096
Min VRAM: 14 GB
GGUF: Yes

Fine-tuned Llama 2 for Hindi with strong bilingual performance.

Sarvam-Translate

Sarvam AI · India

Open

Params: 500M
Context: 512
Min VRAM: 1 GB
GGUF: No

Efficient translation model for multiple Indic languages.

Sutra-Light

Tata Consultancy Services (TCS) · India

Restricted

Params: 1B
Context: 4,096
Min VRAM: 2 GB
GGUF: No

Lightweight enterprise model with broad Indic language support.

Notable vendors

AI4BharatAbhinandAdya AICoRover.aiCommunityIIT BombayIIT MadrasKrutrim AI (Ola)ParamAISarvam AITII (Technology Innovation Institute)Tata Consultancy Services (TCS)

Run benchmarks

llama.cpp, MLPerf, and inference leaderboards.

VRAM calculator

Will this model fit on your GPU?

Hardware guides

Local LLM setup, homelab, mini-PC builds.

Methodology

How we benchmark, verify ASINs, and source data.

Beginner's guide

New to local AI? Start here.

Sourcing & commerce disclosure

Model metadata on this page is compiled from each vendor's public HuggingFace repo, model card, or research paper at time of indexing. Performance figures (KMMLU, MMLU, HAERAE, AraBench, IndicGEM, etc.) link directly to source benchmarks where available; missing or partial values are shown as ", ".

Outbound hardware links (GPUs, mini-PCs, accelerators) are Amazon Associate links using store ID fredoline-20. As an Amazon Associate we earn from qualifying purchases. Prices and availability shown elsewhere on the site are snapshots, not live quotes — verify on Amazon before purchasing. See About / disclosures for the full policy.