News
Feb 22, 2026
Origen Secures $50 Million Strategic Investment from Bluefive Capital to Scale Abu Dhabi AI Solutions
News
Startups
Artificial Intelligence
Asia
NewDecoded
4 min read
Image by sarvam
Bengaluru-based startup Sarvam AI officially launched Indus on February 20, 2026, marking a significant step toward digital independence for the nation. The platform offers a localized chat interface powered by the Sarvam-105B model, which was built entirely from scratch in India. Currently in limited beta, the tool aims to provide an accurate and efficient AI experience tailored to the unique linguistic and cultural needs of the domestic market.
Unlike global models that often struggle with the nuances of regional speech, Indus natively supports 22 Indian languages and excels at code-switching. This feature allows users to mix Hindi and English naturally, mirroring everyday conversation patterns across the subcontinent. The interface is specifically designed to be accessible, moving beyond simple text input to handle the complexities of local communication.
A core differentiator is the platform's voice-first approach, which prioritizes speech over typing for ease of use. It includes high-fidelity speech-to-text and text-to-speech capabilities optimized for various Indian accents. Additionally, the system includes Sarvam Vision, a proprietary model designed to handle complex Indian scripts in documents and images more effectively than standard global tools according to sarvam.ai.
At its technical core, the Sarvam-105B uses a Mixture-of-Experts architecture to achieve high performance with reduced inference costs. This allows the model to activate only a fraction of its parameters per token, ensuring efficiency without sacrificing reasoning capability. While smaller than some international frontier models, Sarvam reports that its flagship outperforms competitors like China's DeepSeek R1 on specific regional benchmarks.
The launch is a cornerstone of the company's Sovereign AI mission, which advocates for national control over critical digital infrastructure. By building the full technology stack from data centers to the user interface, Sarvam aims to ensure that Indian data remains within the country's borders. This approach is supported by the IndiaAI Mission, which has provided subsidized compute capacity to accelerate the training of these homegrown solutions.
Sarvam is also expanding its reach through key strategic partnerships with hardware manufacturers like HMD and Bosch. A smaller 30B model is being optimized for use in Nokia feature phones to bring voice-AI assistants to non-smartphone users. Furthermore, collaborations with Qualcomm and Yotta Data Services ensure that the models are optimized for both on-device processing and scalable cloud infrastructure.
For now, interested users can access the interface through a limited beta on web and mobile platforms. The company is gradually rolling out access to manage compute capacity, so a waitlist remains in place for some new sign-ups. Feedback from this early phase will be used to refine the model's accuracy and alignment before a wider public release is finalized.
The introduction of Indus signals a shift in the global AI landscape where national sovereignty is becoming as valuable as raw compute power. By moving away from being a mere wrapper for American models, Sarvam AI is proving that mid-sized, specialized models can effectively challenge general-purpose giants in local markets. This development validates India's capacity to own its data and interface layers, potentially inspiring other nations to pursue similar self-reliance in the face of Silicon Valley's dominance.