VibeStudio, an emerging Chennai-based AI lab focused on secure on-premise solutions, has introduced a pruned AI coding model tailored for enterprise environments that operate without internet access. Alongside the launch, the company has also been selected for Intel’s accelerator program, marking an important step toward optimizing its technology for Intel’s upcoming Panther Lake processors. The move offers Indian institutions a homegrown alternative at a time when Chinese open-source AI models dominate the sector.
A Push Toward AI Self-Reliance
The launch comes amid rising concerns around India’s increasing dependence on Chinese AI models such as DeepSeek, GLM, and MiniMax. Industry experts have repeatedly highlighted the long-term risks of relying on foreign technologies, especially for sectors where data security is non-negotiable.
VibeStudio, which operates out of the ITEL Foundation under Padma Shri awardee Professor Ashok Jhunjhunwala, hopes to fill this gap. The company was founded by Arjun Reddy, who previously built GuardianLink, a Kalaari Capital-backed venture.
“We achieved this with 200 dollars worth of compute,” Reddy said, calling attention to the team’s efficient approach. “With 1 million dollars, we can build a well-documented, open-weights model performing close to 90 percent of the Kimi K2 Thinking class. Our aim is to help India build privacy-preserving AI for critical sectors.”
How VibeStudio Pruned a 250B Model
At the heart of the new release is VibeStudio’s proprietary THRIFT process, short for Targeted Hierarchical Reduction for Inference and Fine-Tuning. Using this technique, the company managed to prune the MiniMax-M2 model, a large Chinese Mixture-of-Experts system, from 250 billion to 192 billion parameters without losing its coding performance.
The pruned model is lighter, easier to deploy, and uses far less computing power – qualities that matter for government and banking networks that prefer offline or restricted setups.
The model has already crossed 10,000 downloads on Hugging Face, adding to the company’s total of more than 150,000 downloads across releases. VibeStudio now plans to apply THRIFT to the Kimi K2 Thinking model and reduce its size by 50 percent by December 2025, potentially enabling offline AI use on smaller devices.
Intel Partnership to Boost Local AI Computing
In another major development, Intel has selected VibeStudio for its accelerator program. The collaboration focuses on tuning the startup’s AI suite for Intel’s upcoming Panther Lake processors, built on the advanced 18A process and designed with powerful Neural Processing Units.
The aim is to ensure that VibeStudio’s tools run optimally on hardware that many Indian enterprises will adopt next year. This pairing of local AI models with next-generation chips is expected to strengthen the domestic AI ecosystem significantly.
Early Adoption by Indian Companies
Several Indian businesses have already begun using VibeStudio’s models.
• IppoPay, a fast-growing UPI and fintech company, has deployed the models in secure coding workflows
• GUVI, an HCL-backed ed-tech platform serving more than 3.5 million learners, uses VibeStudio’s tools to support internal automation
Their adoption signals rising interest in secure, network-isolated AI alternatives built within India.
Why Indian Enterprises Need Domestic AI Tools
Recent data cited by Silicon Valley firm Andreessen Horowitz (a16z) suggests that 80 percent of early-stage AI startups depend on Chinese open-source models. For sectors like BFSI, healthcare, and education where the confidentiality of information is crucial this dependency introduces long-term risks.
VibeStudio’s latest release intends to provide a credible, privacy-first alternative that offers the capability of global models without exposing data to external networks.
Frequently Asked Questions (FAQs)
Q1: What is VibeStudio?
A1: VibeStudio is a Chennai-based AI lab that builds secure, on-premise AI systems for banks, enterprises, and government organizations that require strict privacy controls.
Q2: What is the THRIFT method?
A2: THRIFT is VibeStudio’s pruning technique used to shrink large AI models. It cuts down memory use and processing needs while keeping the model’s intelligence largely intact.
Q3: Why has VibeStudio partnered with Intel?
A3: The startup joined Intel’s accelerator program to optimize its models for Intel’s Panther Lake processors, ensuring better performance and power efficiency.
Q4: What is the concern around Chinese AI models?
A4: Chinese models like MiniMax and DeepSeek are powerful, but relying on them creates dependency on foreign technology. For regulated sectors in India, domestic solutions provide stronger control and better data sovereignty.
Q5: Who is currently using VibeStudio’s AI models?
A5: Early adopters include IppoPay, a fintech and UPI services company, and GUVI, an HCL-backed ed-tech platform. Both use VibeStudio’s on-premise models for secure coding and automation.
