Shakti AI Endpoint
The Fast Lane to AI Power,
Without the Roadblocks.
Shakti AI Endpoints deliver GPU-optimised, low-latency model hosting that’s ready for anything - language, image, digital twins, and beyond. Built on the muscle of NVIDIA NIMs, it gives you enterprise-grade security, auto-scaling smarts, and seamless API integration, all wrapped in a pay-per-use model. Seamless MLOps with Shakti AI Endpoints.
Deploy endpoints > Fine-tune models > Run inference > Monitor performance
Built with the Best
Accelerating Multimodal AI Innovation
at Scale with Shakti AI Endpoints.
Transform user interactions by integrating lifelike digital avatars or virtual assistants into your applications, powered by models like Meta’s Llama and NVIDIA Riva.
Unlock AI-driven creativity with models like NVIDIA NeMo and Mixtral. Generate personalised and domain-specific content based on your proprietary data.
With NVIDIA Clara and BioNeMo, streamline biomolecular generation and rapidly explore compounds and molecular structures to accelerate new drug and therapy development.
Using advanced models like NVIDIA Omniverse and Siemens MindSphere, create real-time virtual replicas of your physical assets to test, optimise, and innovate.
Personalise shopping, optimise pricing, and improve inventory accuracy.
Boost network performance, automate support, and enhance user experience.
AI Endpoints Advantage
Optimized Inference Infrastructure for Intelligent Workloads with Shakti AI Endpoints.
NVIDIA NIMs
GPU-optimised models for peak throughput, ultra-low latency, and high concurrency, built for demanding AI workloads.
Built-in Observability & SLA Monitoring
Real-time insights ensure your AI endpoints meet performance and uptime commitments without surprises.
Enterprise-Grade Security
API key protection, encryption, and compliance baked in.
OpenAI-Compatible API
Start running Shakti AI models in your app with just a few lines of familiar OpenAI-style code.
Domain-Ready Models
Tailored solutions for healthcare, automotive, BFSI, gaming, and more.
Optimized Multi-Modal Workloads
Unified inference layer for text, vision, and speech models, enabling cross-modal applications like conversational search and generative media.
TensorRT & Quantization Optimization
Endpoints leverage NVIDIA TensorRT and Quantization to slash latency and inference costs while maintaining accuracy.
Hybrid Inference Deployment
Run workloads seamlessly across on-premises, private cloud, and Shakti Cloud endpoints with unified APIs.
Peak Performance
Unveiling the Secrets of
High-Performance Architecture
- OpenAI-Compatible API
- Enterprise-Grade Security
- NVIDIA-Optimised Models
- Auto-Scaling Infrastructure
- Model Marketplace Access
OpenAI-Compatible API
Shared Endpoints expose OpenAI-compatible APIs, enabling developers to instantly integrate with their existing AI applications. This ensures fast onboarding and migration without the need to rewrite code.
Enterprise-Grade Security
Even in shared environments, data is protected with encryption in transit, strict API authentication, and logical isolation. This allows customers to safely experiment without compromising compliance requirements.
NVIDIA-Optimised Models
Shared Endpoints serve pre-hosted NVIDIA-optimized models for LLMs, Vision, and Speech workloads. Optimizations ensure inference is delivered with consistent performance across multi-tenant environments.
Auto-Scaling Infrastructure
Shared Endpoints are backed by elastic GPU infrastructure that auto-scales based on requests per second (RPS). Rate limits are enforced to ensure fairness across tenants, while maintaining low latency.
Model Marketplace Access
Customers can instantly try from a curated set of pre-integrated models (LLMs, ASR/TTS, Vision) hosted on the platform. Shared Endpoints allow rapid prototyping and proof-of-concept deployments without needing dedicated infrastructure.
Why Shakti Cloud Works for You
Flexible AI Pricing, Optimized for Your Workloads
- Pay-as-you-Go
Llama 3.1 8BVersatile open – source model for general – purpose tasks.2020
| Model Name | Description | 1 M Input Token | 1 M Output Token |
|---|---|---|---|
| Llama 3.1 70B | High – parameter base model for language understanding tasks. | ₹75 | ₹75 |
| Llama 4 Scout 17B 16E Instruct | High – performance language model. | ₹17 | ₹59 |
| Llama 4 Maverick | High – performance language model. | ₹25 | ₹76 |
| DeepSeek R1 Distill Llama 8B | High – performance language model. | ₹8 | ₹8 |
| DeepSeek R1 Distill Llama 70B | High – performance language model. | ₹138 | ₹156 |
| Mistral Large 2 | High – performance language model. | ₹183 | ₹535 |
| Mixtral 8x7B | Mixture of Experts model for scalable multitask inference. | ₹53 | ₹53 |
| Gemma 3 (4B) | Compact model from Google for lightweight NLP tasks. | ₹4 | ₹7 |
| Gemma 3 (12B) | High – performance language model. | ₹9 | ₹12 |
| Gemma 3 (27B) | High – performance language model. | ₹22 | ₹36 |
| Qwen3 30B | High – performance language model. | ₹24 | ₹77 |
| Qwen3 14B | High – performance language model. | ₹36 | ₹128 |
| Qwen3 235B | High – performance language model. | ₹63 | ₹247 |
| Qwen3 32B | High – performance language model. | ₹63 | ₹249 |
| Qwen2.5 VL 72B Instruct AWQ | High – performance language model. | ₹106 | ₹108 |
| Qwen QWQ 32B | High – performance language model. | ₹107 | ₹107 |
| DeepSeek R1 (INT 4) | High – performance language model. | ₹484 | ₹484 |
| DeepSeek V3 (INT 4) | High – performance language model. | ₹299 | ₹299 |
| Mistral Nemo Inferor 12B | High – performance language model. | ₹15 | ₹15 |
| Minimax M1 | High – speed general model optimized for API – first use cases. | ₹41 | ₹188 |