Sarvam AI

Sarvam AI is an Indian artificial-intelligence company headquartered in Bengaluru, Karnataka, focused on building foundation models and applied AI systems for Indian languages and local use cases. The company was founded in August 2023 by Vivek Raghavan and Pratyush Kumar, who had previously been associated with AI4Bharat at the Indian Institute of Technology Madras. Its name is derived from a Sanskrit term associated with “all” or “wholeness,” reflecting the company’s stated interest in broad language access and inclusive AI capabilities. The company’s work centers on large language models trained on multilingual Indian-language data, including code-mixed text. Its technical portfolio has included text-generation models, speech recognition, vision-language systems, and other multimodal technologies. Sarvam has used mixture-of-experts architectures in some of its models, a design approach intended to increase model capacity while activating only a portion of parameters for each token. The company positions its systems for applications such as translation, voice interfaces, document understanding, enterprise automation, and developer services. Sarvam AI obtained early venture backing in 2023. In December of that year, it announced a combined seed and Series A financing round of approximately US$41 million led by Lightspeed Venture Partners, with participation from Peak XV Partners and Khosla Ventures. The company subsequently became involved in India’s public effort to develop domestic AI capabilities. In 2025, the Ministry of Electronics and Information Technology selected Sarvam as one of the organizations contributing to an indigenous foundational-model programme under the IndiaAI Mission. The arrangement included access to government-supported computing infrastructure for model training. The company has also pursued institutional and ecosystem collaborations. In March 2025, the Unique Identification Authority of India announced a collaboration with Sarvam AI concerning AI-enabled voice interactions and multilingual support for Aadhaar-related services. Sarvam has worked in areas related to Indian-language datasets and speech research, including activity connected with AI4Bharat and academic institutions. It has also participated in open-source AI initiatives and was listed among Indian members of the AI Alliance. In February 2026, Sarvam introduced the Sarvam-30B and Sarvam-105B foundational models at the AI Impact Summit in New Delhi. Sarvam-30B was described as a 30-billion-parameter mixture-of-experts model with a 32,000-token context window, while Sarvam-105B was described as a 105-billion-parameter model aimed at more demanding reasoning and enterprise workloads, with a 128,000-token context window. A beta version of Sarvam-105B was subsequently made available under the name Indus through web and mobile channels. The two principal models were also released as open-source models under the Apache License 2.0, with weights made available through Hugging Face and AIKosh. Beyond text models, Sarvam has developed Saaras V3, a speech-to-text system supporting multiple Indian languages, and Sarvam Vision, a vision-language system intended for document understanding and optical character recognition in Indian scripts. The company has described Sarvam Kaze as an AI-enabled wearable-glasses product designed to listen, interpret, and capture aspects of a user’s visual environment while supporting voice interaction in more than ten Indian languages. It has also launched a startup programme offering selected early-stage companies API credits, engineering assistance, and access to production infrastructure for multilingual AI applications. Sarvam AI’s overall strategy combines proprietary model development, open model distribution, public-sector collaboration, and developer or enterprise infrastructure. Its differentiation is primarily geographic and linguistic: rather than treating Indian languages as a secondary feature of g…

Sarvam AI was established in August 2023 in Bengaluru by Vivek Raghavan and Pratyush Kumar. Both founders had prior associations with AI4Bharat at IIT Madras, an initiative known for work on Indian-language artificial intelligence and language resources. Sarvam was created to develop AI systems that treat India’s linguistic diversity and local institutional requirements as core design considerations. The company’s early activity concentrated on large language models and related infrastructure. Its models use multilingual datasets containing Indian languages and code-mixed text, and some employ mixture-of-experts architectures. This technical direction allows Sarvam to target text generation, reasoning, coding, translation, and enterprise workflows while maintaining a focus on languages and scripts that have historically received less representation in mainstream AI systems. In December 2023, Sarvam announced approximately US$41 million in combined seed and Series A financing. The round was led by Lightspeed Venture Partners and included Peak XV Partners and Khosla Ventures. The funding supported the company’s model-development and infrastructure ambitions as competition intensified among Indian AI startups. Sarvam’s role expanded from private-sector model development into government-supported AI infrastructure. In 2025, India’s Ministry of Electronics and Information Technology selected the company for participation in the IndiaAI Mission’s indigenous foundational-model programme. Under that initiative, Sarvam received access to government-supported computing capacity, including GPUs allocated for model training. The company also entered a collaboration with UIDAI in March 2025 to explore AI-based voice interaction and multilingual support for Aadhaar-related services. These activities connected Sarvam’s language technology with large-scale public-service use cases. In February 2026, Sarvam presented two foundational models, Sarvam-30B and Sarvam-105B, at the AI Impact Summit in New Delhi. Sarvam-30B was characterized as a 30-billion-parameter mixture-of-experts model with a 32,000-token context window. Sarvam-105B was positioned for more complex reasoning and enterprise use, with a 128,000-token context window. A beta version of the larger model was subsequently offered under the name Indus through web and mobile applications. The models were later released openly under the Apache License 2.0, with weights distributed through Hugging Face and AIKosh. The company’s portfolio also includes speech and vision systems. Saaras V3 is a speech-to-text model designed to support multiple Indian languages, while Sarvam Vision is intended for document understanding and optical character recognition in Indian scripts. Sarvam has additionally described Sarvam Kaze, an AI wearable-glasses concept capable of voice interaction, visual capture, and support for more than ten Indian languages. Its Sarvam Startup Program extends the company’s platform strategy by giving selected startups credits, engineering access, and production resources for multilingual applications. Sarvam AI’s development has therefore followed several parallel tracks: foundation-model research, open-source distribution, speech and vision products, government collaboration, and developer ecosystem support. The company’s central strategic proposition is that AI systems for India should be trained, evaluated, and deployed with attention to the country’s languages, scripts, speech varieties, and public and commercial service environments.

1. 2026Sarvam-30B and Sarvam-105B introduced

   Sarvam unveiled two foundational language models at the AI Impact Summit in New Delhi.

   Source
2. 2026Models released under the Apache License

   Sarvam made the weights of its 30B and 105B models available through Hugging Face and AIKosh under the Apache License 2.0.

   Source
3. 2025Selected for the IndiaAI Mission

   Sarvam was selected to develop an indigenous foundational model under India’s IndiaAI Mission and received access to supported computing infrastructure.

   Source
4. 2025UIDAI collaboration announced

   UIDAI announced collaboration with Sarvam AI on multilingual and voice-based capabilities for Aadhaar-related services.

   Source
5. 2023Sarvam AI founded

   Vivek Raghavan and Pratyush Kumar founded Sarvam AI in August 2023 in Bengaluru.

   Source
6. 2023Seed and Series A financing announced

   The company announced approximately US$41 million in combined seed and Series A financing led by Lightspeed Venture Partners.

   Source

An India-focused AI platform and model developer emphasizing multilingual Indian-language capabilities, sovereign or locally relevant AI infrastructure, and applications spanning text, speech, vision, enterprise software, and public services.

• Sarvam-30BFoundational large language model2026

  A 30-billion-parameter language model designed around Indian-language and multilingual tasks. It uses a mixture-of-experts approach and was described as activating approximately one billion parameters per token, with a 32,000-token context window. The model supports general generation and reasoning use cases as well as mathematics, coding, translation, and enterprise applications. Its weights were released under the Apache License 2.0 through public model repositories.

• Sarvam-105BFoundational large language model2026

  A 105-billion-parameter foundational model intended for more demanding reasoning and enterprise workloads. The model was described as using a mixture-of-experts architecture with roughly 9 billion active parameters per token and a 128,000-token context window. Sarvam positioned it for advanced multilingual tasks, coding, mathematics, and complex application development. It was released openly under the Apache License 2.0.

• IndusAI assistant and model application2026

  A beta presentation of the Sarvam-105B model made available through web and mobile applications. Indus provides an end-user interface for interacting with Sarvam’s large language model capabilities and is intended to demonstrate multilingual and reasoning-oriented use cases.

• Saaras V3Speech-to-text model

  A multilingual speech-recognition system developed for Indian-language use cases. Saaras V3 is intended to convert spoken language into text and support voice interfaces, transcription, translation-related workflows, and other applications where speech access is important.

• Sarvam VisionVision-language model

  A vision-language system focused on document understanding and optical character recognition in Indian scripts. The product is intended to help software interpret visual documents and extract or reason over information in multilingual Indian-language contexts.

• Sarvam KazeAI wearable device2026

  An AI-enabled wearable-glasses concept designed to listen, interpret, and capture aspects of the user’s visual environment in real time. Sarvam has described support for voice interaction in more than ten Indian languages, including possible translation and contextual-assistance use cases.

• 2026Sarvam Startup Program

  India

  Sarvam AI launched a programme for selected early-stage companies building multilingual applications in areas such as speech, translation, and language models. Participants receive API credits for six to twelve months, priority engineering support, and access to production infrastructure scaled to their needs.

  Outcome. The programme expanded Sarvam’s developer and startup ecosystem strategy; broader participation or commercial outcomes were not specified in the supplied sources.

  Source

• 2026Open release of Sarvam-30B and Sarvam-105BProduct launch

  Sarvam AI was developing large language models for Indian-language, reasoning, coding, mathematics, and enterprise applications.

  What changed. The company released the two models under the Apache License 2.0 and distributed their weights through Hugging Face and AIKosh.

  Aftermath. The release made Sarvam’s principal language models available to developers and organizations for experimentation and deployment, subject to the applicable license terms.

  Source
• 2025Participation in the IndiaAI MissionStrategy

  India sought to strengthen domestic AI infrastructure and develop foundational models suited to national requirements.

  What changed. Sarvam AI accepted selection by the Ministry of Electronics and Information Technology to contribute to an indigenous foundational-model programme and use supported computing resources.

  Aftermath. The collaboration placed Sarvam among the Indian companies contributing to a government-backed sovereign-AI effort.

  Source
• 2025Multilingual Aadhaar service collaborationStrategy

  UIDAI sought to improve voice interaction and multilingual access for Aadhaar-related services.

  What changed. UIDAI announced a collaboration with Sarvam AI to explore AI-based voice interactions and multilingual support.

  Aftermath. The collaboration connected Sarvam’s language and speech technology with a large-scale Indian public-service environment.

  Source

• 2026Sarvam AI introduces Sarvam-30B and Sarvam-105B

  Sarvam AI presented two large language models at the AI Impact Summit in New Delhi. The models were designed for Indian-language tasks and broader reasoning, coding, mathematics, and enterprise applications.

  Product launchProduct generation
  Source
• 2026Sarvam AI releases open-source language models

  Sarvam AI released Sarvam-30B and Sarvam-105B under the Apache License 2.0, making model weights available through Hugging Face and AIKosh.

  Product launch
  Source
• 2026Sarvam AI launches Sarvam Startup Program

  The programme offers selected startups API credits, priority engineering support, and production infrastructure for multilingual AI applications.

  Campaign
  Source
• 2025Sarvam AI selected for IndiaAI Mission foundational-model programme

  India’s Ministry of Electronics and Information Technology selected Sarvam AI to contribute to an indigenous foundational-model initiative and provided access to government-supported computing resources.

  Other
  Source
• 2025UIDAI collaborates with Sarvam AI on multilingual voice services

  The Unique Identification Authority of India announced work with Sarvam AI focused on AI-enabled voice interaction and multilingual support for Aadhaar-related services.

  Other
  Source
• 2023Sarvam AI announces approximately $41 million seed and Series A financing

  Sarvam AI announced a combined seed and Series A financing round led by Lightspeed Venture Partners, with participation from Peak XV Partners and Khosla Ventures.

  M&AOther
  Source

1. Sarvam AI
2. Sarvam AI official website