India-based AI startup KiteFishAI has introduced NanoI1, a 155-million-parameter language model designed for enterprise applications that require local deployment and greater control over sensitive data.
The model has been trained from scratch on approximately one trillion tokens using eight NVIDIA H100 GPUs in Gurugram. With a file size of 311 MB, NanoI1 is designed to run on laptops and local servers, as well as within enterprise-owned infrastructure and Indian data centres.
Released under the Apache 2.0 licence, the model’s open weights allow developers and organisations to adapt it for specific applications. Its intended use cases include KYC document extraction, complaint classification, customer support ticket routing and short-document processing.
KiteFishAI said NanoI1 is designed for specialised enterprise workflows rather than as a general-purpose alternative to larger language models. The model has a 2,048-token context window, supports vLLM serving and is available in Base and Instruct versions. It supports English, Hindi and eight other Indian languages, although multilingual capabilities are limited.
In benchmark evaluations conducted by KiteFishAI under the same test settings, NanoI1 recorded a 9.1% score on HumanEval, compared with 2.4% for Google’s Gemma-3-270M. On EvalPlus, it scored 4.3% against 1.2%, while its GSM8K mathematics score was 4%, compared with 1.8% for Gemma-3-270M. These figures are based on the company’s own evaluations.
“For many enterprises, the challenge is not simply adopting AI, but deploying it in a way that meets their privacy, infrastructure and cost requirements,” said Anuj Gupta, Founder and CEO of KiteFishAI.
“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,” Gupta said.
The launch comes as enterprises explore smaller models for specific applications where deploying larger AI systems may involve higher infrastructure costs or data-sharing considerations. KiteFishAI is positioning NanoI1 as an option for organisations seeking to run selected AI workloads within their own infrastructure.