VC & Startups/Nov 18 edition/Reporting & analysis
FundingEcosystems

FundingRounds, leads & terms

Bajaj Finance and Pennant Technologies invest $4M in Soket AI, the lab behind India's EKA-1 model

Bajaj Finance and Pennant Technologies have each put about $2M into Soket AI, the Bengaluru lab behind the 26.3B-parameter EKA-1 model unveiled under IndiaAI Mission. No valuation was disclosed, and the model has no published benchmarks.

From the Inc42 article of November 18, 2026 and cited sources: Bajaj Finance invested roughly $2M through its Finserv Intelligence initiative, and Pennant Technologies, a company Bajaj Finance has invested in, added about $2M. [1] [5] [6]

SOURCES[1] Inc42: Soket AI raises $4 Mn from Bajaj Finserv inc42.com[5] Business Standard: Bajaj Finserv Intelligence, Pennant… business-standard.com[6] Daily Prabhat: Pennant invests USD 2 million in Soket AI dailyprabhat.com

AI-assisted, not human-edited. Methodology Corrections Who owns this site

KEY FACTS
What happened
Bajaj Finance invested roughly $2M through its Finserv Intelligence initiative, and Pennant Technologies, a company Bajaj Finance has invested in, added about $2M. [1] [5] [6]
Original source
Inc42 article, independent perspective
Brief by
AT Pulse research desk AI-assisted; publisher and contact on the desk page
When
November 18, 2026 article published
Evidence
12 sources across 9 sites

Reach: not measured yet

Illustration from Inc42 Media: Bajaj Finance and Pennant Technologies invest $4M in Soket AI, the lab behind India's EKA-1 model
Image: Inc42 Media — Original article ↗
THE CORE IDEAS4 TAKEAWAYS
01

Bajaj Finance invested roughly $2M through its Finserv Intelligence initiative, and Pennant Technologies, a company Bajaj Finance has invested in, added about $2M. The total is about $4M. The valuation and the size of the investors' stakes were not disclosed. [1] [5] [6]

02

EKA-1 is a sparse mixture-of-experts model with 52 layers and 128 experts, trained from scratch on about 7.5T tokens. It targets reasoning, code and agentic tasks in finance and cybersecurity. Its active parameter count, technical report, model card and benchmark results have not been published. [3] [4] [7]

03

The round is small next to Soket's original goal of building a 120B-parameter open model. That model work still depends on 1,536 H100 GPUs subsidised by IndiaAI, and EKA-1 is a much smaller step toward that target. [2] [10] [11]

04

The investors are also likely customers. Pennant is adding agentic AI to its banking software and wants models that banks can run on their own infrastructure, which matches how EKA-1 is pitched for on-premises and air-gapped use. [6] [12]

WHY IT MATTERS

The reported facts are a $4M strategic round with no disclosed price and a model with no published benchmarks.

Read the full assessment

The implication: financial firms are funding domestic models they may deploy themselves, but the round reveals little about how Indian foundation-model labs are valued.

Executive brief

Bajaj Finance and Pennant Technologies, a company Bajaj Finance has invested in, have put about $4M into Soket AI. Each put in roughly $2M. Soket is the Bengaluru lab behind EKA-1, a 26.3B-parameter model it calls sovereign, which IndiaAI Mission officials unveiled in early November 2026. The surprising part is the round's size. $4M is small for a lab that set out to build a 120B-parameter open model, and that work depends on government-subsidised compute. Valuation and stake were not disclosed. EKA-1 was released without published benchmark results. In practice, this deal links a strategic financial-sector investor to a lab whose model has not yet been evaluated in public.

What changed and event timeline

  1. IndiaAI picks Soket

    The IT Ministry selected Soket AI Labs, Gnani.ai and Gan.ai from 506 applicants. Soket's goal was India's first open-source 120B-parameter foundation model.

  2. 3–4 Nov 2026: EKA-1 unveiled

    IndiaAI Mission and MeitY unveiled EKA-1 in New Delhi, alongside ZenteiQ's Dhīr models. Soket's post gives 3 November; Inc42 reports 4 November.

  3. $4M strategic round

    Bajaj Finance invested about $2M through its Finserv Intelligence initiative, and Pennant invested about $2M. No valuation was disclosed.

Capabilities and access

  • Model: EKA-1, 26.3B parameters, trained from scratch on about 7.5T tokens. Inc42
  • Intended uses: maths, code generation, reasoning and agentic tasks, aimed at banking, financial services and cybersecurity.
  • Languages: reports conflict. Inc42 says the 22 scheduled Indian languages plus English. Soket and Pennant say 100+ languages.
Read the full section
  • Model: EKA-1, 26.3B parameters, trained from scratch on about 7.5T tokens. Inc42
  • Intended uses: maths, code generation, reasoning and agentic tasks, aimed at banking, financial services and cybersecurity.
  • Languages: reports conflict. Inc42 says the 22 scheduled Indian languages plus English. Soket and Pennant say 100+ languages. Soket AI on X; Daily Prabhat
  • Access: the government said the models can be offered through its AIKosh platform. No licence terms have been published. Let's Data Science
  • Other release: Soket has also released Loop, an open-source harness for running long tasks with agents.

Technical analysis for researchers and developers

  • Architecture: described as an "open, sparse hybrid" model. Soket's post lists 52 layers and 128 experts, which points to a mixture-of-experts design.
  • Data: a dataset called EKA pretraining Indic corpus v1.1 is listed on AIKosh. AIKosh
  • Gaps for reproducibility: no technical report, tokenizer details, evaluation suite or inference requirements have been published.
Read the full section
  • Architecture: described as an "open, sparse hybrid" model. Soket's post lists 52 layers and 128 experts, which points to a mixture-of-experts design. The active parameter count and the meaning of "hybrid" (for example, attention mixed with another layer type) are not documented. Soket AI on X; Let's Data Science
  • Data: a dataset called EKA pretraining Indic corpus v1.1 is listed on AIKosh. AIKosh
  • Gaps for reproducibility: no technical report, tokenizer details, evaluation suite or inference requirements have been published. Without the active parameter count, serving costs cannot be estimated.

Claims and evidence

  • Round size and split ($2M + $2M): announced by the companies and reported by Inc42 and Business Standard. No regulatory filing confirming it was found.
  • A sponsored post also names Finvolve and India Accelerator as co-investors in what it calls a "seed round" of undisclosed size.
  • Valuation: not disclosed.
Read the full section
  • Round size and split ($2M + $2M): announced by the companies and reported by Inc42 and Business Standard. No regulatory filing confirming it was found.
  • Investors: Bajaj Finance and Pennant. A sponsored post also names Finvolve and India Accelerator as co-investors in what it calls a "seed round" of undisclosed size. The mainstream reports do not mention them. Business Upturn (brand post)
  • Valuation: not disclosed.
  • Model quality: no benchmark results have been published, so performance claims cannot be checked. Let's Data Science
  • Investor comments: Rajeev Jain, Vice Chairman and MD of Bajaj Finance, and Pennant's founders Pradeep Varma and Rama Raju are speaking about a company they invested in.

Context and prior work

  • Compute: Soket received access to 1,536 NVIDIA H100 GPUs through IndiaAI. AIM
  • Scale: EKA-1 is far smaller than the 120B target, and Soket says it plans to scale further. Inc42
  • Other Indian models: ZenteiQ released Dhīr 2B and 5B as lightweight models that can run on devices.
Read the full section
  • Compute: Soket received access to 1,536 NVIDIA H100 GPUs through IndiaAI. AIM
  • Scale: EKA-1 is far smaller than the 120B target, and Soket says it plans to scale further. Inc42
  • Other Indian models: ZenteiQ released Dhīr 2B and 5B as lightweight models that can run on devices. Sarvam AI released Prism, built on its Sarvam 105B model. Inc42

Limitations, safety and contested findings

  • No benchmarks, safety evaluation or model card have been published. Inc42
  • Language coverage is reported two ways: 22 Indian languages plus English, or 100+ languages.
  • The unveiling date is reported two ways: 3 or 4 November.
Read the full section
  • No benchmarks, safety evaluation or model card have been published. Inc42
  • Language coverage is reported two ways: 22 Indian languages plus English, or 100+ languages.
  • The unveiling date is reported two ways: 3 or 4 November.
  • The investor list differs between mainstream reporting and the sponsored post.
  • The round is strategic money from investors who are also likely customers. Pennant is building agentic AI into its banking software and wants models that banks can run on their own infrastructure. Daily Prabhat

Business and practitioner implications

  • For banks and other regulated firms: EKA-1 is pitched for on-premises and air-gapped deployment.
  • For Soket: $4M will not pay for frontier-scale training.
  • For investors: strategic financial-sector money is replacing a valuation-setting venture round.
Read the full section
  • For banks and other regulated firms: EKA-1 is pitched for on-premises and air-gapped deployment. That fits data-residency requirements, but buyers should run their own tests before relying on it. madebyagents
  • For Soket: $4M will not pay for frontier-scale training. Soket's model work still depends on IndiaAI-subsidised compute, while the new capital goes to applied research and enterprise deployments in finance, cybersecurity and defence.
  • For investors: strategic financial-sector money is replacing a valuation-setting venture round. Without a disclosed price, the round reveals little about how the market values Indian foundation-model labs.
FOLLOW THE EVIDENCE

The source trail.

Sources (12)
A LITTLE LESS NOISE. A LOT MORE CONTEXT.

Stay curious.
Follow the evidence.

Independent perspectives, the original sources, and room for the questions that don't have easy answers.

How we build the brief
Connect with us

Find us where you already read.