Product Launches & Tech Releases ⏱ 4 min. Read (729 words)

KiteFishAI Introduces NanoI1, a Compact AI Model for Private and Cost-Efficient Enterprise Deployments

India-developed 155M-parameter language model trained on approximately 1 trillion tokens aims to help enterprises deploy AI on their own infrastructure while enhancing data privacy and reducing costs

- Advertisement -

Key Highlights:

  • KiteFishAI has launched NanoI1, a compact 155-million-parameter language model designed for private, efficient and cost-effective enterprise AI deployments.
  • The model was trained from scratch on approximately 1 trillion tokens using 8 NVIDIA H100 GPUs in Gurgaon and has a compact file size of just 311 MB.
  • NanoI1 supports deployment on laptops, local servers and enterprise-owned infrastructure, helping organizations maintain greater control over sensitive data.
  • Released under the Apache 2.0 open-source licence, NanoI1 enables developers and enterprises to customize the model for specialized use cases.
  • The model supports enterprise applications including KYC document extraction, complaint classification, customer support ticket routing and short-document processing.
  • NanoI1 is available in Base and Instruct variants and supports English, Hindi and eight other Indian languages.

India, November 17, 2026:
As enterprises increasingly adopt artificial intelligence across business operations, concerns around data privacy, infrastructure costs and dependence on third-party AI platforms continue to pose challenges, particularly for organizations handling sensitive customer information.

Addressing these concerns, KiteFishAI has introduced NanoI1, a compact language model designed to enable private and cost-efficient AI deployments on enterprise-controlled infrastructure.

The company positions NanoI1 as a practical alternative for specialized enterprise workflows where organizations require tighter control over data, infrastructure and deployment environments.

- Advertisement -

Built and Trained in India
NanoI1 was trained entirely from scratch using approximately 1 trillion tokens on 8 NVIDIA H100 GPUs in Gurgaon.

Key specifications include:

  • 155 million parameters
  • 311 MB model size
  • 2,048-token context window
  • Support for vLLM serving
  • Base and Instruct (chat) variants

Its compact design allows deployment across:

  • Enterprise-owned servers
  • On-premise environments
  • Local infrastructure
  • Laptops and edge computing environments

This enables businesses to explore AI use cases without routinely transmitting sensitive information to external AI providers.

Enterprise-Focused AI Applications
NanoI1 has been designed specifically for focused business workflows rather than general-purpose AI tasks.

Potential use cases include:

  • KYC document extraction
  • Complaint classification
  • Customer support ticket routing
  • Workflow automation
  • Short-document processing
  • Structured data applications

The model supports deployment within Indian data centres and enterprise-controlled environments, providing organizations with greater flexibility over compliance and data governance requirements.

Open-Weight Architecture
NanoI1’s weights have been released under the Apache 2.0 licence, enabling businesses, developers and startups to modify and fine-tune the model for specific industry requirements.

According to KiteFishAI, this approach helps organizations retain greater control over:

  • Infrastructure choices
  • Data privacy
  • Model customization
  • Deployment architecture

The company believes open-weight models can accelerate innovation while reducing barriers to enterprise AI adoption.

Benchmark Performance
KiteFishAI stated that NanoI1 delivers competitive performance across several benchmark evaluations despite its relatively small size.

In company-conducted evaluations under identical test settings:

HumanEval Coding Benchmark

  • NanoI1: 9.1%
  • Gemma-3-270M: 2.4%

EvalPlus Benchmark

  • NanoI1: 4.3%
  • Gemma-3-270M: 1.2%

GSM8K Mathematics Benchmark

  • NanoI1: 4.0%
  • Gemma-3-270M: 1.8%

According to the company, these results are notable given that Gemma-3-270M contains nearly twice the number of parameters.

Multilingual Support
NanoI1 supports:

  • English
  • Hindi
  • Eight additional Indian languages

The company said the model also offers limited multilingual capabilities beyond its primary language coverage, making it suitable for a range of domestic enterprise applications.

Anuj Gupta, Founder & CEO, KiteFishAI, said: “For many enterprises, the challenge is not simply adopting AI, but deploying it in a way that meets their privacy, infrastructure and cost requirements.”

He added: “With NanoI1, we wanted to demonstrate that a compact model, developed from scratch in India, can deliver meaningful performance on specialised tasks while giving organisations greater control over their data and deployment.”

Commenting on the broader opportunity for compact AI models, Gupta said: “We believe smaller, purpose-built models can play an important role in making AI more accessible and practical for businesses, particularly in regulated industries.”

Supporting India’s AI Ecosystem
Through NanoI1, KiteFishAI aims to contribute to India’s growing AI development ecosystem by making compact, efficient and open-weight models accessible to:

  • Enterprises
  • Startups
  • Developers
  • Research teams

The company emphasizes that NanoI1 is designed as a foundation model for specialized enterprise applications rather than as a direct replacement for large-scale general-purpose language models.

About KiteFishAI
KiteFishAI is an India-based artificial intelligence company founded by Anuj Gupta and focused on developing compact, high-performance AI models.

Its work spans:

  • Small Language Models (SLMs)
  • Embedding Models
  • Large Language Models (LLMs)
  • Retrieval-Augmented Generation (RAG)
  • AI Agents
  • Scientific AI

The company focuses on advancing India’s AI capabilities through the development of efficient models and supporting infrastructure.

Its Nano-Em1-0.6B-v2.1 model ranked among the top-performing models on the MTEB leaderboard in 2026, reflecting the company’s emphasis on efficient AI engineering and model development.

Join the conversation

- Advertisement -
What did you think of this story?

Topics in this story

ai model india AI startup India Anuj Gupta Apache 2.0 Artificial Intelligence Artificial Intelligence Startup compact ai model enterprise AI enterprise ai deployment enterprise language model EvalPlus Gemma-3-270M generative AI GSM8K HumanEval kitefishai kyc document extraction ai language model machine learning MTEB Nano-Em1-0.6B-v2.1 nanoi1 NVIDIA H100 on premise ai open source ai open weight ai model small language model vLLM

Amit Verma ✓

🌐 StartupConsultant.in

With over 24 years of hands-on experience, Amit Verma, Founder of Startup Success Stories, is dedicated to empowering startups and established businesses with practical, result-driven strategies. His core expertise lies in lean startup methodologies, business strategy, and fundraising; helping ventures…

Founder Spotlight

Building a Breakout Indian Startup?

Share your journey and get featured across India's fastest-growing venture media ecosystem.

Submit Your Story →

Related Stories