Two days after unveiling a 105-billion-parameter language model at a summit in New Delhi, Sarvam did what every AI startup eventually must: it released a consumer app. On February 20, 2026, the Bengaluru-based company pushed Indus—a voice-first, multilingual chatbot—into public beta. The timing was deliberate, perhaps more so than the launch itself suggested.
India has become the second-largest market for generative AI tools. Sam Altman has said ChatGPT now logs 100 million weekly active users in the country, according to a Times of India op-ed reported by TechCrunch on February 15. Anthropic reports India accounts for 5.8% of Claude's global usage, trailing only the United States, per TechCrunch's launch coverage. OpenAI and Google have already staked their claims. Sarvam is betting that a chatbot designed explicitly for Indian languages, running on servers physically located in India, can still carve out meaningful territory.
Whether that bet pays off depends on more than just the technology.
The Model: Big, Multilingual, and Sovereign
Indus runs on Sarvam-105B, the flagship model announced February 18 at the India AI Impact Summit. The architecture is mixture-of-experts, a design that activates different subnetworks depending on the task. All 22 scheduled Indian languages are supported, according to press coverage from outlets including Business Standard, though Sarvam has not yet published a formal technical report with benchmarks or detailed specifications.
Context windows are reported at up to 128,000 tokens for the 105B model—comparable to what OpenAI and Anthropic offer, at least on paper. A smaller sibling, Sarvam-30B, handles real-time chat with roughly 32,000 tokens of context. Both models are designed for what the company calls "sovereign deployment," meaning they run on infrastructure inside India. It's a pitch that aligns with government priorities around data localization, though the commercial appeal to consumers is less clear.
The app's knowledge cutoff is June 2025, based on testing by Business Today. That's nearly a year behind ChatGPT's, a gap that could prove meaningful in fast-moving domains like policy, tech news, or global events.
Voice-First, With the Usual Beta Caveats
Indus is available on iOS, Android, and web (indus.sarvam.ai). At launch, access was gated by a waitlist and restricted to India. The iOS version 1.6 update on March 5 quietly added the ability to sign in "outside India" using Google, Apple, or Microsoft accounts—a signal that geographic restrictions are loosening, even if the app remains tuned for Indian users. The Android listing shows over 100,000 downloads in the first two weeks, a respectable start for a capacity-constrained rollout.
The app accepts typed and spoken queries across languages. Early reviewers have praised its ability to handle mid-conversation code-switching—English to Hindi, or toggling between regional dialects—without the friction that often plagues multilingual interfaces. Voice input returns both text and audio responses, a design choice that reflects Sarvam's broader ambitions in speech technology.
Document and image uploads are supported, letting users ask questions about PDFs or photos. There's also a "Web, simplified" feature that scans the internet for direct answers, though the extent and reliability of real-time search integration remains unclear from early reviews.
The beta carries predictable limitations. Users cannot delete individual chats without wiping their entire account. There's no toggle to disable the model's "reasoning" mode, and some early adopters report missing features—image upload unavailable on certain clients, occasional bugs in voice processing. Whether these are temporary gaps or permanent design choices is hard to say. Cofounder Pratyush Kumar acknowledged compute constraints in posts around the release, and the March 5 updates to both iOS and Android added dark mode and improved new-chat UX, suggesting the team is iterating quickly.
More Than Just a Chatbot
Indus is the public face of a broader stack. Sarvam's ambition isn't simply to build another chatbot—it's constructing an end-to-end AI platform tuned for Indian languages and deployable on Indian infrastructure. On February 15, the company announced Sarvam Akshar, a document intelligence workbench that uses a 3-billion-parameter vision model for layout-aware OCR and grounded reasoning across English and all 22 Indian languages. A February 14 blog post detailed Sarvam Edge, which focuses on on-device inference for compact speech-to-text, text-to-speech, and translation models targeting edge hardware.
The partnerships suggest Sarvam is thinking beyond smartphones. HMD, the Nokia-brand licensee, is working with the company to bring AI features to feature phones—a segment that still matters in India. Bosch has signed on for automotive AI applications. A January 13 memorandum of understanding with the Tamil Nadu government commits to a Rs 10,000 crore "Sovereign AI Park," and a collaboration with Qualcomm is targeting wearable devices. Sarvam is reportedly planning a smartglasses product called "Kaze," though details remain scant.
Funding, Pricing, and the OpenAI Shadow
Sarvam has raised $41 million to date, according to TechCrunch's February 20 launch coverage, from Lightspeed, Peak XV, and Khosla Ventures. (A separate TechCrunch story on February 18 cited total funding of "more than $50 million," though the discrepancy likely reflects timing or inclusion of additional strategic commitments. The company's Series A of $41 million was announced in December 2023.)
The app is free during beta. No premium tiers have been announced, which leaves the business model an open question. When pricing does arrive, it will test how much users value localized language support and sovereign hosting against the polish, ecosystem lock-in, and sheer scale of ChatGPT or Gemini.
That's not a trivial question. The 100 million weekly active ChatGPT users in India represent a formidable moat. OpenAI's rapid product iteration—GPT-4 Turbo, DALL·E 3 integration, voice mode improvements—sets a high bar for feature parity, and Google's Gemini benefits from deep integration with Android and Google Search. Sarvam's bet is that voice-first multilingual access, combined with local infrastructure and superior Indic language support, will resonate with users who feel underserved by English-first tools.
A Market Already Crowded
Sarvam isn't the only Indian company chasing this opportunity. Krutrim, backed by Ola founder Bhavish Aggarwal, is building its own Indic language model. CoRover.ai has been deploying conversational AI for government services. Globally, Meta's Llama models—open weights, multilingual, and increasingly capable—offer a free alternative that developers can fine-tune for local contexts.
The competitive landscape is dense, and it's not clear yet whether there's room for a "third way" between open-source models and the hyperscale incumbents. Sarvam's reported performance on Indic OCR and speech tasks positions it well for document-heavy workflows and voice-driven use cases where ChatGPT's English-first design lags. But the app's beta limitations—no granular chat management, staggered feature rollout, and a knowledge cutoff nearly a year behind ChatGPT's—leave room for competitors to close the gap.
What Happens Next
For now, Indus is live, iterating fast, and betting that local infrastructure and language depth matter more than global polish. The app's emphasis on voice and multilingual fluidity addresses real user needs, particularly in a country where linguistic diversity is a feature, not a bug, of daily life.
But building a better mousetrap isn't enough if the world has already adopted someone else's mousetrap. The next six months will clarify whether Sarvam's sovereign AI pitch resonates with users who are already deeply engaged with global tools, or whether the gravitational pull of OpenAI's scale and Google's distribution proves too strong to overcome.
In the end, technology is only part of the equation. Execution, speed, and the willingness to iterate in public will matter just as much. Sarvam has made a strong opening move. Whether it can sustain momentum in a market dominated by companies with far deeper pockets and far wider reach is the question that will define 2026 for the company—and perhaps for the broader Indian AI ecosystem.
