A team from The International Institute of Information Technology, Hyderabad (IIIT-H) has introduced Patram-7B-InstructIndia’s first vision-language foundational model designed specifically for complex document understanding. This landmark achievement is part of the Bharatgen initiative, a government-supported program to develop India-centric Multimodal Large Language Models, funded by the Department of Science and Technology (DST).
Patram-7B-Instruct is a 7-billion parameter AI model trained on a large, diverse corpus of Indian documents. It can analyze scanned or photographed documents and respond accurately to natural language instructions, making it a versatile tool for varied applications across sectors. Despite its relatively compact size, Patram surpasses larger international models such as DeepSeek-VL-2 on prominent benchmarks like DocVQA and VisualMRC. It also performs strongly on Patram-Bencha custom benchmark designed to reflect real-world Indian document challenges.
The model was officially launched on June 2 at the BharatGen National Summit in New Delhi by Shri Jitendra SinghHon’ble Minister of State for Science and Technology, in the presence of DST Secretary Prof. Abhay karandikarMeitY officials, and other dignitaries. Prof. P. J. NarayananDirector of IIIT Hyderabad, highlighted Patram’s potential to integrate multimodal intelligence from text, speech, and images to power advanced applications.
Developed in just five months by IIIT Hyderabad’s team of alumni and student interns, with support from TiH-IoT IIT Bombay, Patram’s release as an open-source model on platforms like Hugging Face and MeitY’s AIKosh aims to democratize access to cutting-edge AI technologies.
Alongside Patram, the team also launched Docbodha generative AI suite tailored for Indic document intelligence across governance, education, legal, and business sectors.
This breakthrough reinforces India’s commitment to building inclusive, homegrown AI infrastructure in line with national initiatives such as Digital India and Atmanirbhar Bharat.
