In a major victory for India’s sovereign artificial intelligence ecosystem, Bengaluru-based startup Sarvam AI has announced a massive 67% price reduction for its flagship document intelligence platform, Sarvam Vision.

The aggressive price correction follows an exponential surge in enterprise and developer adoption. Since its initial rollout in February 2026, the vision-language model has been used to process and digitize more than 35 million document pages. Rather than pocketing the resulting financial margins, Sarvam AI is passing its back-end infrastructure savings directly on to end-users.

1. The Realignment: Breaking Down the Token Economics

The price adjustment makes large-scale optical character recognition (OCR) and document understanding significantly more affordable for Indian enterprises:

• The Price Drop: The transactional cost to process a document through the Sarvam Vision API has been slashed from ₹1.5 per page down to just ₹0.5 per page.
• The Target Verticals: This 67% drop sharply lowers operating overheads for organizations managing massive physical paperwork mountains, such as financial institutions running automated KYC, healthcare providers digitizing medical records, and public sector archives processing legacy manuscripts.

2. Behind the Stack: Reworking the Sovereign Infrastructure

According to technical updates shared by the company, the price drop wasn’t a temporary promotional subsidy. Instead, it was driven by a complete technical optimization of their deployment stack as client data volumes began to skyrocket.

To handle scale efficiently, Sarvam’s engineering teams overhauled several aspects of their serving architecture:

• Kernel Optimization: They developed custom, optimized inference kernels tailored specifically for State-Space model architecture.
• Intelligent Batching: The platform integrated smarter, page-level computational batching algorithms to prevent processing bottlenecks.
• Sovereign Cloud Utilization: Engineers squeezed significantly higher hardware utilization rates out of the startup’s localized, domestic sovereign cloud infrastructure.

The resulting full-stack structural improvements allowed the model to run at a fraction of its launch cost, enabling the company to lean into what co-founder Pratyush Kumar describes as the “deflationary world of AI itself”.

3. The Local Advantage: Outperforming Global Giants

Launched alongside its massive 30-billion and 105-billion parameter foundation models earlier this year, Sarvam Vision is a compact, highly specialized 3-billion parameter model built natively for India’s unique multilingual requirements. It features comprehensive out-of-the-box support for all 22 official Indian languages.

While global tech giants build massive, general-purpose models hosted in foreign data centers, Sarvam’s hyper-focused, localized optimization has proven to be both highly accurate and highly economical.

By combining top-tier benchmark accuracy with aggressive INR-denominated pricing that avoids the complex 18% GST reverse charges tied to foreign APIs, Sarvam is systematically removing the financial barriers to population-scale AI deployment across India.