Mumbai, February 4: Ola CEO Bhavish Aggarwal announced 'Krutrim AI Labs' today in India along with key announcements for all the models, including Krutrim 2, Krutrim 1, Chitrarth 1, Dhwani 1, Vyakhyarth 1 and Krutrim Translate 1. Bhavish Aggarwal introduced Krutrim AI Lab as India's frontier AI research lab dedicated to developing artificial intelligence for Indian infrastructure. The platform was announced to further help exceptional AI talent with the aim to develop world-class AI models, strengthening India's position as a global leader in artificial intelligence.

Krutrim AI Lab will further help India's talent tap into key research areas such as Efficient AI, Multilingual AI, Multimodal AI, Personalised AI, Embodied AI, and Data-Centric AI. With this announcement, Ola CEO Bhavish Aggarwal released five major AI model categories and what they intend to offer to Indian citizens. Check out all the latest models from the Ola Krutrim platform here. Krutrim AI Lab: Ola CEO Bhavish Aggarwal Releases Open-Source AI Models with Focus on 'Developing AI for India'.

Krutrim 2 and Krutrim 1 LLMs

Bhavish Aggarwal announced the new version, Krutrim 2, in India, which is the successor of Krutrim 1 LLM introduced in January 2024. Krutrim 1 large language model was trained on 7B (7 billion) parameters, which delivered good performance on Indic benchmarks. The new Krutrim 2 LLM is built on the Mistral-NeMo architecture. It is trained on 12B parameters, making it a powerful model. Krutrim-2 was trained on a diverse dataset, including English, 22 Indian languages, code, math, literature, and high-quality synthetic content, supporting a 128K token context window.

Chitrarth 1 - Vision Language Model

Chitrarth 1 is a VLM or Vision Language Model that comes with LLM integration. Ola Krutrim said that this model was trained on a multilingual, multilingual image-text data set with 10 prominent Indian languages, including Hindi, English, Telugu, Bengali, Gujarati, Tamil, Marathi, Kannada, Malayalam, Odia, and Assamese.

Dhwani 1- India's first Speech Language Model

Dhwani-1 is called India's first end-to-end speech LLM and is powered by Krutrim-1. Dhwani-1 directly understands speech without needing a separate ASR model, avoiding ASR errors typically occurring in speech-to-text translation. The model supports translation between English and multiple Indic languages, including Bengali, Gujarati, Hindi, Kannada, Malayalam, Marathi, Tamil, and Telugu. As part of the release, its speech-to-text translation capabilities are being open-sourced.

Vyakhyarth 1 - Indic Embedding model

Vyakhyarth-1 is an Indic embedding model designed for semantic textual similarity, search, clustering, and classification across over 100 languages. It is better than NLP models that struggle with cross-lingual understanding What Is OpenAI Deep Research AI Agent? Know About AI Agent Launched in ChatGPT for Multi-Step Research on Internet for Complex Tasks; Check How To Use It..

Krutrim Translate 1 - Text-To-Text Translation Model

Krutrim Translate 1 is a text-to-text translation model that helps get results in the chosen Indic languages. It has low latency as it underwent various architecture explorations for training and distillation. Krutrim Translate 1 supports multiple Indian languages, including Telugu, Gujarati, Bengali, Hindi, Marathi, Malayalam, Kannada, and Tamil. 

(The above story first appeared on LatestLY on Feb 04, 2025 03:50 PM IST. For more news and updates on politics, world, sports, entertainment and lifestyle, log on to our website latestly.com).