Bengaluru-based Soket AI, a frontier AI research lab building indigenous large language models under the government’s IndiaAI Mission, has raised nearly $4 million from Bajaj Finance and Pennant Technologies.
Bajaj Finance invested about $2 million through Bajaj Finserv Intelligence, the Bajaj Finserv group’s applied research and innovation initiative. Pennant Technologies, a lending software company in which Bajaj Finance holds a stake, put in another $2 million in the same round.
India Accelerator and its investment platform Finvolve said they had also invested an undisclosed amount in what they described as Soket’s seed round.
Soket AI has not disclosed its valuation.
The funding will support Soket AI’s work on large language models focused on coding, mathematics and reasoning, and on applying them to enterprise use, particularly in financial services.
Why a lender wants its own AI model
Financial services companies are testing AI in lending, customer service, risk assessment and cybersecurity. Using these systems also raises questions about where sensitive customer data goes and who controls the infrastructure the models run on.
Soket AI’s pitch is that its models can be run on an organisation’s own infrastructure, including on-premises environments, and that they can be trained and deployed at lower cost. Bajaj Finance said it wants models that deliver results at sustainable cost while running inside its own systems.
“At Bajaj Finance, FinAI is reshaping how we build, operate and scale our business. The next phase of AI will be defined by outcomes delivered at sustainable cost. Soket AI’s efficient reasoning-led models can make advanced AI economically viable for financial services, while allowing us to deploy it securely within our own infrastructure, with full control over data,” said Rajeev Jain, Vice-Chairman and MD of Bajaj Finance.
Sanjiv Bajaj, Chairman and MD of Bajaj Finserv, said that as AI becomes part of critical information infrastructure, the margin for error must be minimal.
“Through Bajaj Finserv Intelligence, we are investing in indigenous research that meets these requirements while applying advanced AI to real-world business challenges. Our investment in Soket AI reflects our commitment to strengthening India’s sovereign AI capabilities and building scalable solutions for India and the world,” he said.
Abhishek Upperwal, founder and CEO of Soket AI, said the startup was “delighted to have the backing of Bajaj Finance Limited”.
Pennant said the partnership would allow its teams to deepen research into AI agents and sovereign language models that regulated institutions can deploy while retaining control over their data and decisions.
Soket AI and Pennant also plan to jointly develop banking-specific language models and AI agents for know-your-customer (KYC) checks, onboarding, collections, compliance and fraud detection.
Bajaj Finserv Intelligence invests in startups, works with academic institutions and employs in-house specialists, and plans over a five-to-10-year horizon to develop technology that can scale, according to the group.
What Soket AI has built
Founded in 2019 by Abhishek Upperwal, Soket AI focuses on “frontier AI for work where a wrong answer has a cost”, with applications across cybersecurity, defence, finance, banking, legal work and agriculture.
In 2024, the startup released Pragna-1B, an open-source multilingual language model with 1.25 billion parameters, which it said was trained from scratch.
In May 2025, the IndiaAI Mission picked Soket AI, along with Gnani AI and Gan AI, to build foundation models, following Sarvam AI’s selection in April that year. Soket was chosen to develop what was described as India’s first open-source 120-billion-parameter foundation model, aimed at sectors such as defence, healthcare and education. The government had received 506 proposals for foundation models by then, according to reports.
The first model from that work arrived on 3 November this year. Soket’s EKA-1, built under its Project EKA, was unveiled in New Delhi alongside ZenteiQ AiTech Innovations’ Dhīr 2B and Dhīr 5B, with IT secretary S Krishnan present.
According to the IT ministry, EKA-1 has 26.3 billion parameters and was trained on nearly 7.5 trillion tokens. It is an open, sparse hybrid language model that supports all 22 scheduled Indian languages and English, and is designed for mathematics, code generation, reasoning and agentic applications.
The ministry said subsidised AI compute and support for research expenses under the mission helped the startups build their models.
“With the support of the IndiaAI Mission, we have developed an advanced AI model specifically tailored to India’s unique banking, financial services and cybersecurity landscape,” Upperwal said at the launch.
Soket has also released Loop, an open-source agent framework for running long tasks, and runs an inference platform that serves EKA and selected open models in on-premises and air-gapped setups.
Soket plans frontier-scale models in the range of hundreds of billions to a trillion parameters, supporting more than 100 languages, including Indian and Global South languages, as well as programming languages.
The Bajaj group’s link with Pennant goes back to October 2023, when Bajaj Finance said it had agreed to acquire up to a 26% stake in the company for Rs 267.50 crore, on a fully diluted basis, to strengthen its technology infrastructure.
Pennant, incorporated in 2005, provides software products and technology services to banks and financial institutions, including AI-powered lending platforms.


/indianstartupnews/media/media_files/2026/06/17/desktop-728-x-90-1-2026-06-17-17-20-35.png)

