GPU inference in India: Private AI and the shared API

Teams searching for GPU inference in India usually want one of three things: GPUs of their own, a provider to run their own model or fine-tune, or models served from India. unoblox covers these in different ways. For dedicated GPUs it offers Private AI, where we run GPUs for you. For everything else it is a shared, per-token API.

Private AI: dedicated GPUs, run for you

Not every workload belongs on shared endpoints. With Private AI, unoblox runs dedicated GPUs for a client, so your capacity is yours and you don't queue behind other customers' traffic. It isn't a self-serve checkbox. Each engagement is scoped with our team, who will talk through your model, your volume and your requirements and tell you plainly what we can set up.

Talk to the team: see Private AI or write to us through the contact page and pick Sales. Scope, pricing and terms are agreed with you directly rather than listed here.

The shared API and the other routes

1. A shared per-token API to the 65 models in the live model list. You use one OpenAI-compatible endpoint and one ub-gw- key, and you're billed in rupees per token. Every listed rate already includes unoblox's 5% platform fee, and you are charged exactly the listed rate.

2. A few open models served from unoblox's own GPUs in India. These are the only models in the live model list that run in India:

Model	Model id	Input (₹ / 1M)	Output (₹ / 1M)
Qwen3.8-27B	qwen/qwen3.8-27b	₹0	₹0
Gemma 4 26B A4B Instruct	google/gemma-4-26b-a4b-it	₹0	₹0
DeepSeek-OCR	deepseek-ai/deepseek-ocr	₹1.00	₹1.00

Qwen3.8-27B and Gemma 4 26B A4B are free, with a fair-use limit of 10 requests per minute per workspace (see Freemium access). DeepSeek-OCR reads documents and images and returns text. It is also tagged Freemium: it's free while your wallet balance is at or above the freemium minimum (₹1,000 by default), and below that it bills at ₹1.00 per 1M tokens for input and output. Every other model in the shared catalog runs at an upstream provider outside India. For those, what's India-specific is rupee billing, a GST invoice for each top-up, and one endpoint.

3. A provider route for teams that run their own GPUs. If you already serve a catalog model on your own infrastructure through a public, OpenAI-compatible endpoint (for example SGLang or vLLM behind HTTPS), you can apply to list it as a provider. Endpoints must pass correctness and load checks before they take traffic. Once admitted, an endpoint can be routed marketplace traffic, not just your own, subject to eligibility, price and capacity. Payouts depend on verification and contracts. See Serve a model and For providers.

4. Bring your own key. If you have your own account with a supported upstream provider, unoblox can route to it on your credentials for a 2% fee instead of the token price. See Bring your own key.

Calling a model served from India

from openai import OpenAI
client = OpenAI(base_url="https://api.unoblox.ai/v1", api_key="ub-gw-…")
response = client.chat.completions.create(
    model="qwen/qwen3.8-27b",
    messages=[{"role": "user", "content": "Extract the invoice number and total from this text."}],
)

Billing

Usage is paid from a prepaid rupee wallet. GST at 18% is added when you top up (pay ₹118 and ₹100 of credit lands in your wallet), and one GST tax invoice is issued for each top-up; individual requests carry no separate invoice.

Frequently asked questions

Can unoblox host my own or fine-tuned model? Talk to the team about Private AI. Dedicated GPUs run for you are the place to discuss it, and we will say plainly what is possible. If you would rather host it yourself on a public OpenAI-compatible endpoint serving a catalog model, the provider route above may fit.

Which models in the shared API run in India? Qwen3.8-27B, Gemma 4 26B A4B and DeepSeek-OCR, served from unoblox's own GPUs. No other model in the live model list runs in India.

Is there dedicated or reserved capacity? Yes, through Private AI: dedicated GPUs run for a client, arranged with our team. The shared per-token API, by contrast, runs on shared endpoints with per-key and per-workspace rate limits.

Who do I talk to about Private AI or a requirement that isn't covered here? Private AI is the starting point, or use the contact page and pick Sales. We'll tell you plainly whether we can help.

Get started in rupees → https://unoblox.ai/sign-in