Easten MediatechEasten Mediatech
← All Insights
Artificial Intelligence · National ·

India Releases New Multilingual AI Models: What Users Should Test

The IndiaAI Mission has introduced EKA-1 and the Dhir 2B and 5B models. Their Indian-language and offline ambitions are notable, but teams should test real tasks before making performance claims.

Conceptual computing core connected to abstract multilingual data paths
Original conceptual illustration; it does not show an actual model interface or benchmark.

What was released

India's Electronics and IT Ministry announced three models under the IndiaAI Mission on 3 November.

Soket developed EKA-1; ZenteiQ developed Dhir 2B and Dhir 5B.

The ministry says EKA-1 covers all 22 scheduled Indian languages alongside English.

It describes the Dhir models as multilingual assistants designed for secure or low-connectivity settings.

Why language support matters

Many Indian services must work across languages, scripts and regional ways of asking the same question.

A model's language list is a starting point, not proof that every dialect or specialist task works equally well.

Public services and businesses should test real user questions rather than relying on a launch description.

A useful evaluation checklist

Test the exact languages, scripts and domain terms your customers use most often.

Measure factual accuracy, refusal behavior, latency and cost on tasks your team already understands.

Ask whether sensitive documents stay on a controlled system and who can access the logs.

Check accessibility and fallback to a human when an answer is uncertain or potentially harmful.

What is not established yet

The announcement does not independently prove that these models outperform alternatives on every task.

Marketing or government teams should not promise perfect translation, reasoning or privacy without evidence.

Transparent pilots and published evaluations will matter more than parameter counts alone.

IndiaAI MissionEKA-1DhirIndian languages

Sources

↑