Indic AI Scorecard

Which model for an Indian-language feature, at what cost?

Loading this month's results…

Quality against cost

One dot per model. Up is better quality; left is cheaper per 1,000 answers (log scale). Bars are 95% ranges: overlapping bars mean a tie. Models worth using sit top-left.

Leaderboard

Failure gallery

Real answers from this run that a code check caught. These are the mistakes that break products.

Answer explorer

Method and limits