← Back to Home
Voice & Local AI
Speech & Language Leaderboards

These leaderboards show how well today's AI models understand and translate Pakistan's languages: Urdu, Punjabi, Pashto, Sindhi and Balochi.

Think of them as a public scorecard. Every AI model takes the same standard tests, known as benchmarks, and is ranked by how accurately it performs two tasks:

Most AI tools are built and tested mainly in English and often perform poorly in local languages. Measuring this openly shows which models are ready for real-world use, such as voice-based health tools for mothers with limited literacy, and where more work is needed. The tests use publicly available datasets (Common Voice, FLEURS, FLORES-200 and IN22-Gen), so anyone can check and reproduce the results.

Having trouble viewing it? Open the leaderboard in a new tab →