Ola Krutrim Announces AI Lab Launch To Mixed Reactions From Users

Ola founder Bhavish Aggarwal on Tuesday announced a slew of launches in its AI portfolio. The biggest of all was a Rs 2,000 crore ($230 million) investment into his AI startup Ola Krutrim to start an AI lab. The lab seeks to help in building an open-source AI ecosystem by fostering research, its website says.

Apart from this, the company also launched the following:

• Krutrim2: A multilingual Large Language Model (LLM) with 12B parameter, optimized for English and 22 Indic languages, and built on the Mistral-NeMo architecture with a 128K token context window. This is the second model by the company, after it launched Krutrim1 in 2024.
• Chitrarth: A multilingual Vision Language Model (VLM) trained primarily on multilingual image-text data and is designed to work across 10 prominent Indian languages, including Hindi, Bengali, Telugu, Tamil, Marathi, Gujarati, Kannada, Malayalam, Odia, and Assamese, as well as English.
• Dhwani: End-to-end trained speech LLM, based on Krutrim-1 LLM. The company has open-sourced the speech-to-text translation capabilities of this model. It supports translation between Indic Languages and English. The supported Indic languages are Bengali, Gujarati, Hindi, Kannada, Malayalam, Marathi, Tamil, and Telugu.
• Vikhyarth: Sentence-transformer model designed for semantic textual similarity, search, clustering, and classification across over 100 languages. This model is trained primarily on multilingual data and is designed to work across 10 prominent Indian languages, including Hindi, Bengali, Gujarati, Kannada, Malayalam, Marathi, Sanskrit, Tamil, Telugu as well as English.
• Krutrim Translate: This model translates the input text into one of the chosen Indic languages. It supports English, Bengali, Hindi, Kannada, Marathi, Malayalam, Gujarati, Telugu, and Tamil.

Scoring AI models

Apart from these models, the company has also come up with its own benchmark—BharatBench—to score the performance of its AI models. It said this was because the existing benchmarks “might miss the nuances of Indian languages and cultural contexts,” with most models being primarily in English, Chinese or European languages.

Additionally, the announcement the partnership between Krutrim and Nvidia to deploy the latter’s Blackwell-based GB200 GPU and promised to turn it into a supercomputer.

The reception of the launches:

The model, like its predecessor, is receiving mixed reactions. Some highlighted how well the model was able to understand as well as respond in regional languages like Hindi. Elsewhere it is being flagged for having some very basic problems. One of the users took to X where the model was not able to solve basic logical reasoning queries.

Some others pointed out the announcements held a level of opaqueness to them as the company had not released all the research papers or documentation to back up its claims. Moreover, one of the users hinted that the model has censorship tendencies similar to those for which DeepSeek has been criticized, a model that Krutrim seems to have imitated in its step to open-source its models.

Ola’s focus on the cultural context:

“The motivation behind this model is to ensure that datasets are ‘for our country, of our country, and for our citizens,’ fostering inclusivity and equitable AI advancements,” Krutrim says about its model Chitrarth.

This is a line of thought all too familiar to anyone who has been following Ola or Krutrim’s past announcements or Aggarwal’s comments. The company has repeatedly positioned itself as the bastion of India’s efforts to achieve dominance in the AI industry. Within this positioning, its argument of leveraging the integration of “cultural context” and accurate representation in Indic AI models has been the most recurring one.

At MediaNama’s PrivacyNama discussion in 2024, Senior Director for Foundational LLM Models at Ola Krutrim, Rajkiran Panuganti, had noted: “We always complain that the Indian perspectives are not represented accurately. So, when we say that, it also means that we are not providing the data in an accurate manner and we are allowing stereotypes to be there. So sovereignty should not be seen just as protecting yourself for your economic benefit, but also you want to enable foreign players also to represent you in a way that we want to.” You can read more of his comments here.

Read More: