Six weeks after we opened the waitlist, IndieRouter is in open beta. DeepSeek V4 Flash (0731) is now serving traffic from our own GPUs in India, and every new account starts with ₹500 of free inference credit to put against your real workload.
If you have been waiting to test a model whose tokens never leave Indian soil, this is it.
What’s live today
DeepSeek V4 Flash (0731) is our flagship open-weight chat model, served on our own GPUs with speculative decoding and automatic prefix caching. It carries a 256K context window and is built for chat, coding, agents, extraction, and structured JSON. It is the same open-weight model you would otherwise be paying dollar rates for from another continent, only here the tokens never leave India.
You get the full OpenAI API surface, same request shapes, same streaming, same function-calling and tool semantics. Your existing SDK code does not change: point base_url at https://api.indierouter.ai/v1 and the model parameter at deepseek-v4-flash, and the rest of your codebase stays exactly as it is.
Every request takes one path, and it stays in-country the whole way:
Flat pricing, in rupees, 24/7
DeepSeek’s own API bills peak rates, roughly 2x its off-peak price, during Indian working hours, which is exactly when you need it most. We do not.
- ₹30/M input tokens
- ₹1/M on cache hits (prefix caching is automatic, so keep stable system prompts up front)
- ₹95/M output tokens
No peak-hour surge, no dollar billing, no FX line on your finance team’s spreadsheet. A prepaid wallet topped up via UPI AutoPay, and a GST invoice on every charge so your accountant claims input credit cleanly.
Sovereignty by architecture, not by addendum
We keep saying “DPDP-compliant by architecture” because it means something specific: there is no offshore path in our routing layer to misroute to. Prompts and outputs are processed and discarded in-country, with zero retention, nothing stored, nothing trainable. For most providers, residency is a clause in a contract that a config change can violate. Ours is a property of where the compute lives.
Start in two minutes
- Sign up at app.indierouter.ai/register
- Verify your email and your ₹500 credit lands instantly
- We approve beta accounts in waves. You will be in within hours
- Create an API key and point your OpenAI SDK at
api.indierouter.ai/v1
That ₹500 is enough to run real agent workloads against a 256K context window and feel the latency difference for yourself. In-region round trips run around 25 ms, versus roughly 240 ms to us-east-1.
What’s next
The beta is not a destination, it is the starting line. Three things are already in motion:
- Qwen3.8-27B, a 27B vision-language model with a 256K context window, arriving with the next GPU expansion
- OCR via PaddleOCR, for document extraction on forms, KYC, and structured text, billed per page
- Embeddings and reranking, plus voice, to round out the RAG and multimodal stack
Each of these lands in the catalogue with a published card and a rate when it ships, with no aliases, no silent swaps, and no roadmap dressed up as a product. We would rather tell you it is coming than pretend it is already here.
We built IndieRouter for Indian developers and startups tired of paying dollar prices with peak-hour multipliers for tokens served from another continent. Same open-weight model, Indian infrastructure, rupee pricing.
Sign up at app.indierouter.ai/register and see you in the beta.