Generative AI in Urdu: Who Is Building for Pakistan's Native Language? Urdu is spoken by roughly 230 million people, yet it makes up a sliver of the text the world's AI models are trained on. A student's side project, a telecom-backed government MOU, and a two-decade-old university lab are each trying to close that gap — on their own.

Abstract

In January 2026, a Pakistani student studying in the United States, Taimoor Hassan, released Qalb — a large language model trained on 1.97 billion Urdu tokens and pitched as the world's largest LLM built exclusively for the language. Coverage called it Pakistan's own "Urdu ChatGPT," and for a country where most generative AI tools still think in English before translating into Urdu, the milestone was real.

But Qalb was not built in a vacuum, and it is not alone. A government-backed consortium had already signed on to build a national Urdu model over a year earlier, and a university language lab in Lahore has been quietly producing the speech and text tools that any Urdu AI system depends on since long before "generative AI" was a household phrase. The question worth asking is not whether Pakistan is building for Urdu — it clearly is — but whether these efforts add up to something bigger than the sum of their parts.