Sarvam AI launches India-first models to challenge OpenAI, Gemini with multilingual capabilities

Indian AI lab Sarvam's new models are a major bet on the viability of open-source AI | TechCrunch

Indian AI lab Sarvam on Tuesday unveiled a new generation of large language models, as it bets that smaller, efficient open-source AI models will be able to grab some market share away from more expensive systems offered by its much larger U.S. and Chinese rivals. The launch, announced at the India AI Impact Summit in New Delhi, aligns with New Delhi's push to reduce reliance on foreign AI platforms and tailor models to local languages and use cases. Sarvam said the new lineup includes 30-billion and 105-billion parameter models; a text-to-speech model; a speech-to-text model; and a vision model to parse documents. These mark a sharp upgrade from the company's 2-billion-parameter Sarvam 1 model that it released in October 2024. The 30-billion- and 105-billion-parameter models use a mixture-of-experts architecture, which activates only a fraction of their total parameters at a time, significantly reducing computing costs, Sarvam said. The 30B model supports a 32,000-token context window aimed at real-time conversational use, while the larger model offers a 128,000-token window for more complex, multi-step reasoning tasks. Sarvam said the new AI models were trained from scratch rather than fine-tuned on existing open-source systems. The 30B model was pre-trained on about 16 trillion tokens of text, while the 105B model was trained on trillions of tokens spanning multiple Indian languages, it said. The models are designed to support real-time applications, the startup said, including voice-based assistants and chat systems in Indian languages. The startup said the models were trained using computing resources provided under India's government-backed IndiaAI Mission, with infrastructure support from data center operator Yotta and technical support from Nvidia. Sarvam executives said the company plans to take a measured approach to scaling its models, focusing on real-world applications rather than raw size. "We want to be mindful in how we do the scaling," Sarvam co-founder Pratyush Kumar said at the launch. "We don't want to do the scaling mindlessly. We want to understand the tasks which really matter at scale and go and build for them." Sarvam said it plans to open-source the 30B and 105B models, though it did not specify whether the training data or full training code would also be made public. The company also outlined plans to build specialized AI systems, including coding-focused models and enterprise tools under a product it calls Sarvam for Work, and a conversational AI agent platform called Samvaad. Founded in 2023, Sarvam has raised more than $50 million in funding and counts Lightspeed Venture Partners, Khosla Ventures and Peak XV Partners (formerly Sequoia Capital India) among its investors.

Upstart Sarvam Unveils AI Model Customized for India Market

An Indian startup called Sarvam AI is making a push to create a viable competitor for the domestic market, unveiling an artificial-intelligence model that it says is more tailored to the languages and cultures of the world's biggest market than the likes of ChatGPT and Claude. The Bangalore-based company announced two models at a high-profile AI summit in New Delhi, a showcase for Prime Minister Narendra Modi's efforts to make his country a leading player in the emerging technology. Sarvam's models are built to be used through voice commands and are accessible through 22 Indian languages, which the company says will be a competitive advantage in the country of 1.45 billion where the vast majority can't read, write or type in English. "Today we show we can bring our own AI to a billion Indians," said Sarvam co-founder Pratyush Kumar during an event in Delhi. Sarvam also offers what's known as "agentic" AI models that can carry out tasks like coding or meeting planning in large part autonomously and with minimal human intervention. The company says its agents could drive enterprise automation in one of the world's fastest growing economies. The startup's unveiling of India-specific models trained from scratch comes as the AI race between the US and China is intensifying. The Modi government is funding AI accelerators and pushing model makers to launch services so the country, one of the largest reservoirs globally of technical talent, isn't left behind. Sarvam has steep challenges in fending off global competitors. The startup has received more than $50 million in funding, including from Lightspeed Ventures LLC and Khosla Ventures, and was last valued at about $200 million. That's tiny compared with Silicon Valley leaders like OpenAI and Anthropic PBC, last valued at $500 billion and $380 billion respectively. It's also small next to companies like Mistral AI, a French pioneer in the technology that is valued at $13.25 billion and is expanding in India with local languages. Sarvam touts its India-first approach and the security it offers by running its models from inside the country. The startup's models are trained on trillions of Indian data sets, particularly those in Indian languages, making it suitable for real-time deployment at scale in the world's most populous country not just in individual languages but also mixed languages such as Hinglish, a spoken language that blends Hindi and English. Get the Tech Newsletter bundle. Get the Tech Newsletter bundle. Get the Tech Newsletter bundle. Bloomberg's subscriber-only tech newsletters, and full access to all the articles they feature. Bloomberg's subscriber-only tech newsletters, and full access to all the articles they feature. Bloomberg's subscriber-only tech newsletters, and full access to all the articles they feature. Bloomberg may send me offers and promotions. Plus Signed UpPlus Sign UpPlus Sign Up By submitting my information, I agree to the Privacy Policy and Terms of Service. In recent benchmark tests, the startup said its models performed with superior accuracy on tasks like optical character recognition for Indian scripts, a breakthrough that could be used for a host of practical utilities in a country where language has been a barrier for digital inclusion. For instance the Sarvam vision model achieved an accuracy of over 84% on document intelligence tasks, eclipsing global models hundreds of times larger in size. "Sovereignty matters much more in AI than building the biggest models," said Sarvam's other co-founder, Vivek Raghavan, at the same event. India is hosting dozens of leading global chief executives, AI founders, country leaders, researchers and policy experts this week as it attempts to position itself as an alternative to the US and China, by leading in "democratizing AI" and taking a cost-efficient, language-diverse route. India's digital infrastructure has relied on foreign technologies for decades so the launch of the Sarvam models is seen as a step toward developing a "sovereign AI" ecosystem within the country. Indian AI startups like Sarvam and BharatGen, which also released a series of India-made models this week, are looking to export their AI systems to other developing economies in the world where neither Chinese nor US models are favored.

This Indian AI startup is claiming victory over Gemini and ChatGPT, here's why

Sarvam AI claims top-tier OCR and multilingual speech performance built specifically for Indian users * Sarvam AI says its Sarvam Vision model beats Gemini and ChatGPT on key OCR benchmarks * The startup focuses on all 22 official Indian languages * Its "sovereign AI" approach aims to build technology tailored specifically to India's needs ChatGPT, Gemini, and other AI chatbots are often very good at reading English and many other languages, but while they can interpret Hindi, they begin to wobble when confronted with more complex scripts or regional nuance among Indian languages. Now, a Bengaluru startup called Sarvam AI is stepping up with models it says can outperform the global rivals when it comes to optical character recognition (OCR) and multilingual speech, particularly when it comes to the tongues of the sub-continent. The Sarvam Vision and Bulbul V3 models are built with India's linguistic complexity in mind. Sarvam Vision can interpret complex tables, understand charts, recognize text in real-world scenes, and generate captions, while Bulbul V3 handles the text-to-speech system. They support all 22 official Indian languages. With 35 voices, Bulbul is able to always sound like a local. As many multilingual users know, the awkwardness of hearing their language pronounced as if it were a distant cousin of English can make someone reluctant to try the technology. A well-trained text-to-speech model that captures rhythm and tone more accurately can make people feel more comfortable using it. And while OCR may not sound glamorous, it quietly powers everything from when you scan a document with your phone, upload a PDF, or digitize an old record. Garbled characters, misread names, and missing context can be a real issue. Sarvam says it will help small business owners and government offices convert records into searchable archives faster and more accurately than otherwise possible. Sovereign AI Sarvam AI calls itself a builder of sovereign AI. The idea is to distinguish itself from foreign platforms. With AI models spreading across government, business, and education, questions of who builds them and whose data they understand matter a lot. Sarvam wants to have tools tailored to India. Sarvam's emergence also nudges a larger conversation about where innovation originates. The AI boom has often been framed as a race among a few dominant players. Yet breakthroughs increasingly come from focused teams solving specific problems. Sarvam appears to have identified a gap in high-quality, language-rich OCR and speech systems for Indian scripts. Of course, benchmarks are snapshots, not guarantees of performance, especially in the real world. The proof of Sarvam's impact will lie in adoption. Plus, if Sarvam's claims hold up, larger AI companies will feel pressure to improve their own support for more languages and scripts. At its best, Sarvam AI's story goes beyond beating Gemini or ChatGPT on a leaderboard and becomes a way of showing technology reflecting the people who use it. If AI is going to shape the next decade of digital life, it will need to speak many languages fluently and read more than just clean English text. Sarvam is betting that attention to detail and cultural specificity can compete with sheer scale. For millions of users who have felt underserved by mainstream AI tools, that bet may feel more like a sure thing. Follow TechRadar on Google News and add us as a preferred source to get our expert news, reviews, and opinion in your feeds. Make sure to click the Follow button! And of course you can also follow TechRadar on TikTok for news, reviews, unboxings in video form, and get regular updates from us on WhatsApp too.

Sarvam unveils two new large language models focused on real-time use, advanced reasoning - The Economic Times

The company said the model is optimised for "efficient thinking", delivering stronger responses while using fewer tokens -- a key factor in reducing inference costs in production environments.Artificial intelligence startup Sarvam on Wednesday launched two new large language models -- Sarvam-30B and Sarvam-105B -- as the Bengaluru-based company expands its push into advanced reasoning and enterprise deployments. The lighter Sarvam-30B is designed for efficient, real-time applications. It supports a context window of up to 32,000 tokens and has been trained on 16 trillion tokens. The company said the model is optimised for "efficient thinking", delivering stronger responses while using fewer tokens -- a key factor in reducing inference costs in production environments. In benchmarks shared at the launch, Sarvam-30B was evaluated against models including Gemma 27B, Mistral-32-24B, OLMo 31.32B, Nemotron-30B, Qwen-30B and GPT-OSS-20B across tasks such as Math500, HumanEval, MBPP, Live Code Bench v6 and MMLU, which test mathematical reasoning and functional correctness. The company indicated competitive performance across general reasoning and coding benchmarks. On the AIME benchmark -- which measures mathematical reasoning under varying compute "thinking budgets" -- Sarvam-30B showed improved performance as compute allocation increased, positioning it alongside other 30B-class reasoning models. Sarvam also introduced Sarvam-105B, a higher-parameter model aimed at more complex reasoning tasks. The model supports a context length of 128,000 tokens and, according to the company, performs on par with several frontier open and closed-source models in its category. The launch marks Sarvam's move into larger-parameter models at a time when Indian AI startups are seeking to build foundational capabilities domestically rather than rely solely on global APIs. As enterprises prioritise cost efficiency, controllability and data residency, mid- to large-parameter open models are emerging as a viable deployment alternative. The Lightspeed and Peak XV Partners-backed startup did not disclose pricing but said both models are built for enterprise use cases including coding assistance, research, analytics and real-time AI agents.

India AI Impact Summit: Sarvam AI unveils 30B and 105B foundational models, aims to take on OpenAI and other giants

The launch marks a significant push toward India-controlled AI infrastructure focused on cost, privacy, and local accessibility. Bengaluru-based AI startup Sarvam AI has introduced two new large language models- Sarvam-30B and Sarvam-105 B. The announcement was made at the India AI Impact Summit 2026, where the company showcased its AI ambitions, including the Kaze smartglasses. These foundational models are large neural networks trained on vast datasets to understand and generate text across a wide range of tasks. Once developed, they can be customised for applications such as coding assistance, research, translation, enterprise analytics, and conversational AI. The startup aims to offer Indian enterprises more control over data, lower dependency on global APIs and stronger support for regional languages. Starting off, Sarvam-30B, the smaller of the two systems, has 30 billion parameters and can support 32,000 tokens in its context window. It is trained on 16 trillion tokens and is intended to strike a balance between performance and efficiency, producing competitive results in reasoning, coding, and problem-solving benchmarks. At the summit, the company demonstrated Vikram, a multilingual chatbot powered by the model. The bot interacted in languages such as Hindi, Punjabi, and Marathi, and it even worked on feature phones. Sarvam-105B, the larger model with 105 billion parameters and a 128,000-token context window, is built for more demanding enterprise workloads. During a live demonstration, it analysed a company's balance sheet in real time, responding to detailed financial queries and contextual prompts. The model is intended for applications that require processing large volumes of structured and unstructured data, including analytics, long-document summarisation, and advanced coding tasks. Interestingly, this is the first time any Indian startup has created large-scale AI models that can compete with global systems in different tasks like reasoning, coding and data analysis, while supporting multiple Indian languages.