19 Feb 2026·Source: The Hindu

3 min

Science & TechnologyNEWS

Sarvam AI Unveils Two New Language Models at AI Summit

Bengaluru-based Sarvam AI launches two open-source language models, Vikram, at AI Impact Summit.

Bengaluru-based Sarvam AI unveiled two language models at the AI Impact Summit. Named Vikram, the models will be open source. Both models had beaten comparable models from around the world at industry benchmarks.

Training a large language model is a computation-and-skill-intensive process, sometimes entailing millions of dollars worth of work put in by graphics processing units (GPUs) working in concert at data centres. A key aim of the models has been to be better at Indian languages, which do not work very well on most AI models owing to the relative scarcity of text content available in Indian languages that could be used to train models. The launch of the much-anticipated models is seen as a milestone for AI development in India.

Key Facts

Sarvam AI, based in Bengaluru, unveiled two AI language models named Vikram.

The models are open source.

The models have reportedly beaten comparable models at industry benchmarks.

The models aim to be better at Indian languages.

UPSC Exam Angles

GS Paper III: Science and Technology - Indigenization of technology and developing new technology.

GS Paper II: Issues relating to development and management of Social Sector/Services relating to Education, Human Resources.

Ethical considerations in AI development and deployment.

In Simple Words

A company in Bengaluru made two new computer programs that understand and speak languages. These programs, called AI language models, are designed to be good at understanding Indian languages.

India Angle

These AI models can help bridge the language gap in India. Imagine a farmer getting weather updates or a shopkeeper using a translation app to serve customers from different states.

For Instance

Think of Google Translate, but specifically improved for Indian languages like Hindi, Tamil, or Bengali. It could help a tourist understand a local shopkeeper or a student access educational material in their mother tongue.

This matters because it can make technology more accessible to people who don't speak English, connecting them to information and opportunities.

AI that speaks your language opens doors for everyone.

Visual Insights

Key Highlights of Sarvam AI's Vikram Models

Key details about the newly launched Vikram language models by Sarvam AI.

Vikram Models: Open Source

More Information

Background

The development of large language models (LLMs) requires significant computational resources and expertise. Training these models involves using powerful graphics processing units (GPUs) in data centers, often costing millions of dollars. The performance of LLMs is heavily influenced by the data they are trained on. Most existing LLMs are trained primarily on English text, leading to suboptimal performance in other languages, especially Indian languages, due to the scarcity of training data. India has been making efforts to boost its capabilities in artificial intelligence. The launch of open-source LLMs tailored for Indian languages is a step towards AI localization. This approach aims to create AI models that are more relevant and effective for the Indian context, considering the diverse linguistic landscape and cultural nuances. Such initiatives are crucial for ensuring that AI benefits reach a wider population and contribute to solving local challenges.

Latest Developments

In recent years, there has been a growing focus on developing AI models that cater to regional languages and specific cultural contexts. Several initiatives have been launched to create datasets and tools for training AI models in Indian languages. These efforts aim to address the limitations of existing models that are primarily trained on English data. The government has also been promoting research and development in AI through various funding schemes and collaborations between academia and industry. Looking ahead, there is an increasing emphasis on ensuring AI ethics and responsible AI development. This includes addressing issues such as bias in AI models, data privacy, and the potential impact of AI on employment. The development of open-source LLMs like Vikram is expected to foster innovation and collaboration in the AI community, while also promoting transparency and accountability in AI development.

Frequently Asked Questions

1. What are the key facts about Sarvam AI's Vikram models relevant for the UPSC Prelims exam?

Sarvam AI, based in Bengaluru, has launched two open-source AI language models named Vikram. These models aim to improve performance in Indian languages and have reportedly outperformed comparable models in industry benchmarks. The models have parameter sizes of 35 billion and 105 billion.

2. What is the significance of Sarvam AI releasing Vikram as open source?

Releasing Vikram as open source allows developers and researchers to freely access, use, modify, and distribute the models. This can accelerate innovation in AI for Indian languages by enabling community contributions and wider adoption. It also reduces dependence on proprietary models.

3. How does the launch of Vikram impact the common citizen in India?

Vikram's focus on Indian languages can improve access to information and services for citizens who are not fluent in English. This can lead to better user experiences in applications like translation, education, and customer service. Improved AI understanding of local languages can also enhance digital inclusion.

4. What are the recent developments related to AI models for Indian languages?

There is a growing focus on developing AI models that cater to regional languages and specific cultural contexts. Several initiatives have been launched to create datasets and tools for training AI models in Indian languages. These efforts aim to address the limitations of existing models primarily trained on English data.

5. What is data scarcity in AI training, and how does it affect Indian languages?

Data scarcity refers to the limited availability of text data in certain languages, including Indian languages, which hinders the training of effective AI models. Because most AI models are trained primarily on English text, their performance on Indian languages is often suboptimal due to the relative scarcity of training data.

6. What are the important numbers associated with Sarvam AI's Vikram models that could be relevant for the UPSC exam?

The parameter model sizes of Vikram are 35 billion and 105 billion. Sarvam AI has received an investment of $50 million. These figures can be useful for remembering the scale and investment behind the project.

7. What are the pros and cons of using open-source AI models like Vikram?

Pros include increased transparency, community-driven improvements, and wider accessibility. Cons can include potential security vulnerabilities, the need for technical expertise to implement and maintain, and the risk of misuse.

8. Why is Sarvam AI's launch of Vikram considered a milestone for AI development in India?

The launch is considered a milestone because it demonstrates India's growing capabilities in developing advanced AI models, particularly for Indian languages. It also promotes open-source AI development, which can accelerate innovation and reduce dependence on foreign technologies.

9. What is the role of GPUs in training large language models like Vikram?

Training large language models requires significant computational power, which is provided by GPUs (graphics processing units). These GPUs work in concert at data centers to process massive amounts of data and perform the complex calculations needed to train the models.

10. Who are the key personalities associated with Sarvam AI and its Vikram models?

Based on available information, Pratyush Kumar and Ashwini Vaishnaw are key personalities associated with Sarvam AI.

Practice Questions (MCQs)

1. Consider the following statements regarding the 'Vikram' language models recently unveiled by Sarvam AI: 1. These models are designed to perform optimally across all global languages. 2. The models are open source. Which of the statements given above is/are correct?

A.1 only
B.2 only
C.Both 1 and 2
D.Neither 1 nor 2

Show Answer

Answer: B

Statement 1 is INCORRECT: The Vikram language models are specifically designed to perform better with Indian languages, which often underperform on globally trained AI models due to data scarcity. They are not designed to perform optimally across all global languages. Statement 2 is CORRECT: The news explicitly states that the Vikram models will be open source, meaning their source code is available for anyone to use, modify, and distribute.

2. Which of the following factors contributes significantly to the underperformance of many AI models when processing Indian languages?

A.Lack of computational power in Indian data centers
B.Relative scarcity of text content available in Indian languages for training models
C.Government restrictions on AI research in India
D.Absence of skilled AI professionals in India

Show Answer

Answer: B

Option B is the correct answer. The news explicitly mentions that a key aim of the Vikram models is to be better at Indian languages, which do not work very well on most AI models owing to the relative scarcity of text content available in Indian languages that could be used to train models.

3. Assertion (A): Training large language models is a computation-and-skill-intensive process. Reason (R): It requires millions of dollars worth of work put in by graphics processing units (GPUs) working in concert at data centres. In the context of the above, which of the following is correct?

A.Both A and R are true and R is the correct explanation of A
B.Both A and R are true but R is NOT the correct explanation of A
C.A is true but R is false
D.A is false but R is true

Show Answer

Answer: A

Both the assertion and the reason are true, and the reason correctly explains the assertion. The news mentions that training a large language model is a computation-and-skill-intensive process, sometimes entailing millions of dollars worth of work put in by graphics processing units (GPUs) working in concert at data centres. This directly supports the assertion.