BSNL Powers Up with AI, Developing Indigenous Small Language Models

BSNL’s Chairman and Managing Director, Robert J. Ravi, has confirmed that the state-run telecom operator is developing in-house small language models (SLMs) to strengthen its AI capabilities. The move is part of BSNL’s broader strategy to leverage artificial intelligence for network management, customer experience, and churn reduction.

Key Highlights

• AI-driven transformation: BSNL is banking on AI to optimize telecom solutions, improve service delivery, and enhance customer satisfaction.
• In-house SLMs: Instead of relying solely on external large language models, BSNL is building its own smaller, domain-specific models tailored for telecom operations.
• Strategic focus areas:
  • Network management – predictive maintenance, traffic optimization.
  • Customer experience – personalized support, faster resolution.
  • Churn reduction – analyzing usage patterns and proactively addressing customer concerns.
• Leadership vision: Ravi emphasized that AI workloads will dominate the telecom sector in the 5G era, making indigenous AI development critical for BSNL’s competitiveness.

This signals BSNL’s intent to position itself not just as a telecom provider but as a tech-driven enterprise, aligning with India’s push for indigenous AI stacks and digital sovereignty.

Here’s a detailed breakdown of how BSNL’s Small Language Models (SLMs) differ from traditional Large Language Models (LLMs), and why this distinction matters in telecom:

Key Differences Between SLMs and LLMs

Aspect	Large Language Models (LLMs)	Small Language Models (SLMs)	Why BSNL Prefers SLMs
Scale	Billions of parameters, massive datasets	Fewer parameters, domain-specific training	Easier to train and deploy for telecom-specific tasks
Resource Needs	Require huge compute power, GPUs, cloud infrastructure	Lightweight, can run on local servers or edge devices	Cost-effective for a state-run telco with limited budgets
Speed & Efficiency	Slower inference, high latency	Faster response times, optimized for real-time tasks	Critical for customer support and network management
Generalization	Broad knowledge across domains	Narrow focus on telecom operations	Better accuracy for BSNL’s internal use cases
Deployment	Cloud-heavy, centralized	Edge-friendly, can be embedded in telecom infrastructure	Supports 5G rollout and localized AI workloads
Cost	Expensive to train and maintain	Lower training and operational costs	Aligns with BSNL’s need for affordable innovation

Why This Matters for BSNL

• Telecom-specific optimization: SLMs can be fine-tuned for tasks like predictive maintenance, call routing, and churn analysis.
• Digital sovereignty: Developing in-house models reduces dependence on foreign AI providers.
• Scalability in 5G era: Lightweight models can be deployed across BSNL’s vast network infrastructure without overwhelming resources.
• Customer experience: Faster, domain-specific AI responses improve support and reduce churn.

In short, BSNL’s SLMs are leaner, cheaper, and more focused, making them a strategic fit for telecom operations, whereas LLMs are powerful but resource-heavy and overly broad for BSNL’s needs.

Infosys Launches Small Language Models Built Using NVIDIA AI

Infosys Unveils Small Language Models – Infosys Topaz BankingSLM and Infosys Topaz ITOpsSLM – Built on NVIDIA AI Stack

The small language models will be integrated into products and services as part of Infosys Topaz offerings to provide enterprises with a foundation to build their specialized models.

Infosys today announced the launch of its small language models – Infosys Topaz BankingSLM and Infosys Topaz ITOpsSLM – built using the powerful NVIDIA AI Stack. The collaboration leverages NVIDIA AI and Infosys Topaz offerings to provide a robust foundation for implementing and scaling enterprise AI. These models are developed as part of the Infosys center of excellence dedicated to NVIDIA technologies and built to help businesses quickly adopt and scale AI.

The small language models utilize general and industry-specific data, enhanced by NVIDIA’s AI Enterprise and NVIDIA AI Foundry in collaboration with Sarvam AI. The models are fine-tuned with Infosys data and integrated into existing offerings, like Infosys Finacle and Infosys Topaz for business and IT operations, creating robust foundational models for industry-specific applications. Infosys also provides these models as services that include pretraining-as-a-service and fine-tuning-as-a-service, to help businesses build their own custom AI models securely, in compliance with industry standards.

As part of the center of excellence, Infosys is working with NVIDIA on NIM™ Agent Blueprints to streamline AI application development and integrate innovations such as the new Digital Human blueprint for customer service, multimodal PDF data extraction and various other use cases for Infosys Topaz offerings. Beyond these, the collaboration extends to digitalization efforts, addressing areas like 3D workflows and digital twins with NVIDIA Omniverse Enterprise, and Infosys Responsible AI suite, using NVIDIA NeMo Guardrails. The center of excellence also unveiled an exclusive AI Experience Zone, featuring the latest capabilities from NVIDIA AI and Infosys Topaz. The zone is designed to foster co-innovation in AI solutions, such as agentic and physical AI use cases, across sectors such as telecommunications, retail, and financial services.

Balakrishna D. R. (Bali), Executive Vice President, Global Services Head, AI and Industry Verticals, Infosys, said, “As we further our enterprise AI journey with NVIDIA, our focus is now on delivering foundational small language models as services for businesses to build on. By integrating the NVIDIA AI stack with Infosys Topaz, we are taking advantage of very advanced enterprise AI capabilities to tackle unique business challenges, enhance operational efficiency, and deliver bespoke solutions that drive business value for our clients. Our dedicated center of excellence ensures continuous innovation and establishes Infosys as a preferred partner for our clients’ AI-powered transformation.”

Jay Puri, Executive Vice President, Worldwide Field Operations, NVIDIA, said, “Generative AI and the recent advancements in agentic and physical AI are ushering in a new era of innovation and productivity for enterprises worldwide. NVIDIA's full-stack AI platform combined with Infosys Topaz empowers businesses to build and deploy custom AI applications that will transform industries, helping businesses unlock their full potential.”

NVIDIA Releases Small Language Model With State-of-the-Art Accuracy

NVIDIA recently introduced the Mistral-NeMo-Minitron 8B, a compact language model that combines state-of-the-art accuracy with efficiency.

The Mistral-NeMo-Minitron 8B is a miniaturized version of the previously released Mistral NeMo 12B model. It has been pruned from 12 billion parameters down to 8 billion, making it more lightweight while maintaining high accuracy.

In its use cases, this model performs exceptionally well across various benchmarks, including language understanding, common sense reasoning, mathematical reasoning, summarization, coding, and generating truthful answers. It's suitable for AI-powered chatbots, virtual assistants, content generators, and educational tools.

Unlike larger language models, the Mistral-NeMo-Minitron 8B can run in real time on workstations and laptops. This makes it easier for organizations with limited resources to deploy generative AI capabilities while optimizing for cost, operational efficiency, and energy use.

Running language models locally on edge devices enhances security since data doesn't need to be transmitted to a server from the edge device.

In summary, this small language model packs a punch in terms of accuracy and efficiency, making it a valuable addition to the AI landscape.

Moreover, Mistral-NeMo-Minitron 8B stands out due to its compact size and impressive accuracy. While GPT-3 by OpenAI is widely known, it has a massive parameter count (175 billion) and requires substantial computational resources. In contrast, the Mistral-NeMo-Minitron 8B achieves competitive performance with just 8 billion parameters, making it more accessible for smaller-scale applications.

BERT (Bidirectional Encoder Representations from Transformers) is another influential model. However, BERT focuses on context-based embeddings rather than generative capabilities. The Mistral-NeMo-Minitron 8B is more versatile, handling both understanding and generation tasks.

Google's Text-to-Text Transfer Transformer (T5) is a powerful model that frames all NLP tasks as text-to-text problems. While T5 is versatile, the Mistral-NeMo-Minitron 8B's efficiency and real-time deployment edge give it an advantage.

In summary, the Mistral-NeMo-Minitron 8B offers a compelling trade-off between accuracy and efficiency, making it an attractive choice for various applications.

Developers can get started with Mistral-NeMo-Minitron 8B packaged as an NVIDIA NIM microservice with a standard application programming interface (API) — or they can download the model from Hugging Face. A downloadable NVIDIA NIM, which can be deployed on any GPU-accelerated system in minutes, will be available soon.

NVIDIA Unveils 1st Digital Human Technologies On-Device Model for Game Characters

NVIDIA has introduced its first digital human technology on-device small language model called Nemotron-4 4B Instruct. This model is designed for role-playing in games and offers leading retrieval-augmented generation and function-calling capabilities.

Nemotron-4 4B Instruct allows game characters to respond more intuitively to players, comprehend instructions, and perform accurate and relevant actions.

The Nemotron-4 4B Instruct model is being showcased in the multiplayer mech game Mecha BREAK, developed by Amazing Seasun Games. By running directly on GeForce RTX AI PCs, it enhances in-game immersion and provides a dynamic gameplay experience.

Players can interact with mechanic non-playable characters (NPCs) that assist in choosing mechanized robots, appearance customization, and battle preparation.

A Small Language Model Purpose-Built for Role-Playing

NVIDIA Nemotron-4 4B Instruct provides better role-play, retrieval-augmented generation and function-calling capabilities, allowing game characters to more intuitively comprehend player instructions, respond to gamers and perform more accurate and relevant actions.

The model is available as an NVIDIA NIM microservice, which provides a streamlined path for developing and deploying generative AI-powered applications. The NIM is optimized for low memory usage, offering faster response times and providing developers a way to take advantage of over 100 million GeForce RTX-powered PCs and laptops.

James, Nvidia's interactive digital human knowledgeable about NVIDIA and our products. James uses a collection of NVIDIA NIMs, NVIDIA ACE, and ElevenLabs digital human technologies to provide natural and immersive responses

Nemotron-4 4B Instruct is part of NVIDIA ACE, a suite of digital human technologies that provide speech, intelligence and animation powered by generative AI. It’s available as a NIM for cloud and on-device deployment by game developers.

NVIDIA ACE and digital human technologies continue to expand their footprint in the gaming industry.

Global game publisher and developer Perfect World Games is advancing its NVIDIA ACE and digital human technology demo, Legends, with new AI-powered vision capabilities. Within the demo, the character Yun Ni can see gamers and identify people and objects in the real world using the computer’s camera powered by ChatGPT-4o, adding an augmented reality layer to the gameplay experience. These capabilities unlock a new level of immersion and accessibility for PC games.

››› download the NIM to begin building game characters powered by generative AI.

Zoho Making SLMs With Up to 20 Billion Parameters: CEO Sridhar Vembu

Zoho, the Software-as-a-Service (SaaS) giant, is developing smaller artificial intelligence (AI) models or better called as Small Language Models (SLMs). According to founder-CEO Sridhar Vembu, these models are based on 7 billion to 20 billion parameters. The Chennai-based company has found that smaller models are better suited for specific domain problems. Additionally, Zoho aims to have its own graphics processing units (GPUs) infrastructure, which is more cost-effective in the long term.

"We are working on models that are based on 7 billion to 20 billion parameters…we are not doing the 500 parameter models as of now. We also want to have our very own graphics processing units (GPUs) infrastructure as that is cheaper in the long term,” Vembu said at CNBC TV18-Moneycontrol's Global AI Conclave.

Notably, Zoho integrates a range of language models (LLMs) within its workflows. These LLMs are used to improve AI output by infusing them with customer and industry-specific data. Zoho essentially plays one LLM against another to achieve better results.

Zoho takes full advantage of Small(er) Language Models (SLMs) to control AI operating costs while maintaining high-quality outputs. By using smaller models with 7 billion to 20 billion parameters, Zoho aims to solve domain-specific problems for its customers.

SLMs like Llama, Mistral, Qwen, Gemma, or Phi3 are designed to be more efficient at focused tasks such as conversation, translation, summarization, and categorization. They offer tailored solutions that are not only cost-effective but also more accessible, allowing for a broader range of applications and innovations.

Zoho's Chief Evangelist, Raju Vegesna, emphasizes that the best AI implementation is when customers don't even notice they're using AI. In other words, the AI seamlessly enhances their experience without being intrusive.

Additionally, Zoho has revealed plans to develop its own extensive language model (LLM), similar to OpenAI's GPT model and Google's PaLM 2. Furthermore, the company is venturing into chipmaking and seeking incentives from the Indian government for this endeavor.