Supporting case
IndiaAI backs models for Indian languages and local deployment
What happened
At a November 3 event in New Delhi, IndiaAI ↗ presented Soket’s EKA1 and ZenteiQ’s Dhīr, backed by subsidized compute and research support. EKA1 is a 26.3-billion-parameter model trained across Indian languages and English. ZenteiQ ↗ describes two smaller Dhīr models designed for local use; it lists a roughly 1.21 GB compressed version of the smaller model and says training used Google Cloud hardware in Mumbai. The model distribution endpoint did not establish that intended users can download the packages on the stated terms.
Dates and implementation stage
Action/stage 2026-11-03; publication 2026-11-03–04; approved 2026-11-06
