08/09/2026
BHASHINI is India's sovereign language AI platform designed to make language technology open, accessible, and scalable for every Indian. At its core sits NHLT (National Hub for Language Technology), a unified, open & interoperable, and scalable ecosystem connecting people, technology, and opportunity for every Indian, backed by open-source models and APIs, production-grade infrastructure, and a mission to serve India's full linguistic diversity.
The platform is already operating at national scale: 22M inferences daily and 9B inferences till date, with response times under 200ms, all optimised to run on fewer than 200 GPUs, a sovereign system built for efficiency as much as scale.
BHASHINI now hosts world-class Indic AI models from Bodhan AI and AI4Bharat, spanning four core capabilities:
Indic-Transcribe โ an open 1.2B-parameter ASR model delivering accurate speech recognition across all 22 constitutional Indian languages, two Hindi dialects, and Bhili and Bhojpuri, handling code-mixed speech with native, romanized, or mixed-script output.
Indic-Translate โ a 4B-parameter document-level translation model covering English and all 22 Eighth Schedule languages bidirectionally, capable of translating whole documents in a single request, in native script, Roman script, or a mix of both.
Indic-Speak โ a text-to-speech engine with 45 voices across 22 Indian languages and 12 scripts, built to read native and Latin scripts together in one sentence, without language tags.
IndicOCR โ a two-stage document parser for English and 22 Indian languages, combining a 33M-parameter layout parser with a 0.8B block-level OCR model.
Together, these position BHASHINI as the connective infrastructure layer for India's language AI ecosystem, sovereign, open, and built to make sure every Indian language is understood natively. As the platform's own tagline puts it: "Transcending the Language Barrier."
