Guide · Updated 22 Oct 2026

Tamil-language AI

About 6.9 crore people in India speak Tamil as their mother tongue, and AI that works well in Tamil is now within reach. IIT Madras's AI4Bharat publishes free Tamil translation, speech and text models; Sarvam AI's models and voice tools support Tamil; and Tamil Nadu plans its own Tamil large language model through the Tamil Virtual Academy. Here is what exists, who makes it and how to try it.

Facts last checked 22 Oct 2026 · 37 sources · 5 min read

Why Tamil needs its own AI work

Census 2011 counted 6,90,26,881 people in India whose mother tongue is Tamil, 5.70% of the population, which makes Tamil the fifth-largest scheduled language. (Census of India 2011, Paper 1 of 2018 (copy hosted by Down To Earth))

General AI models still find Indian languages hard. In MILU, a test of knowledge in 11 Indian languages including Tamil, the best model reached about 72% average accuracy, and models tuned for a single language did only slightly better than random guessing. (MILU paper (arXiv))

Tamil Nadu's own Tamil language model

The revised 2026-27 budget, presented on 5 Aug 2026, said the state will build a Tamil large language model, with the Tamil Virtual Academy leading the work. The stated aim is to strengthen Tamil computing, build language tools, improve translation and widen Tamil use online. No amount or timeline has been published. (DT Next, The News Minute)

The Tamil Virtual Academy was set up by the state on 18 May 2000 and has long listed Tamil computing goals such as spell checkers, speech recognition and machine translation. (IT and Digital Services Department, Tamil Nadu)

Separately, a statement on the January 2026 MoU between Tamil Nadu and Sarvam AI for a Sovereign AI Park in Chennai said the park would build Tamil-first foundation models. (MediaNama)

Our story on the Tamil LLM plan AI policy in Tamil Nadu

Free and open Tamil AI from IIT Madras

AI4Bharat is an IIT Madras initiative that builds open-source language AI for Indian languages, from datasets to models. (AI4Bharat on Hugging Face)

IndicTrans2 translates between English and all 22 scheduled Indian languages, including Tamil. (IndicTrans2 paper (arXiv))

IndicConformer turns speech into text in all 22 official languages, with a Tamil model, under the MIT licence. (AI4Bharat IndicConformer (GitHub))

Indic Parler-TTS reads text aloud, officially supports Tamil, lets you describe the voice you want in words, and is licensed Apache 2.0. (Indic Parler-TTS model card)

IndicBERT is a small model for tasks like sorting and tagging text in 12 languages including Tamil, under the MIT licence. (AI4Bharat IndicBERT)

For people training their own models, IndicCorp v2 has a Tamil text section under CC-0, IndicVoices holds 7,348 hours of speech across 22 languages including Tamil, and IndicVoices-R has about 1,704 hours of voice data for text-to-speech, licensed CC-BY-4.0. (IndicCorp v2 (Hugging Face), IndicVoices paper (arXiv), AI4Bharat IndicVoices-R)

In September 2026 Bodhan AI, an IIT Madras-incubated centre working with AI4Bharat and NVIDIA, released open Indian-language models for speech, translation and reading text from images. The report does not list which languages. (Careers360 (PTI))

Sarvam AI's Tamil models and voice tools

Sarvam AI's model list shows Tamil among the 11 languages of its main chat model, Sarvam-105B. The model has a 128,000-token context window and was released as open source under Apache 2.0 in March 2026. (Sarvam AI docs: models, Sarvam AI docs: Sarvam-105B, Storyboard18)

Sarvam-Translate translates between English and all 22 scheduled languages, including Tamil. Saaras turns Tamil speech into text and can also give an English translation or a version in Roman script. Bulbul speaks Tamil, with more than 30 named voices in version 3. (Sarvam AI docs: Sarvam-Translate, Sarvam AI docs: Saaras, Sarvam AI docs: Bulbul)

Sarvam marks some older models, including Sarvam-M and the Saarika speech model, as deprecated. (Sarvam AI docs: models)

Other models that handle Tamil

Tamil-Llama, by Abhinand Balachandran, adapted Meta's Llama 2 for Tamil in November 2023 by adding 16,000 Tamil tokens, and released its models, data and code. (Tamil-Llama paper (arXiv))

Google lists Tamil among the Gemini web app's languages; Google Cloud offers Tamil text-to-speech through Gemini-TTS and Tamil speech recognition with codes for India, Sri Lanka, Singapore and Malaysia. (Google Gemini Help, Google Cloud: Gemini-TTS, Google Cloud Speech-to-Text reference)

Meta's SeamlessM4T accepts Tamil speech and text and can write Tamil text, but not speak it, under a non-commercial licence. Ola's Krutrim-2 says it works in English and 22 Indian languages, and Tamil is one of 11 languages in its published tests. (SeamlessM4T paper (arXiv), Krutrim tech blog)

BharatGen lists a 150-million-parameter Tamil text-to-speech model on AIKosh under the MIT licence, with restricted access. (AIKosh: BharatGen Tamil TTS)

Our model tracker marks which models say they support Tamil:

Gemini 3.8 FlashGoogle · closed · Gemini API (Google AI Studio, free tier), Google CloudGoogle Cloud docs list Tamil (ta) among the languages all Gemini models understand and respond in.
IndicTrans2AI4Bharat (IIT Madras) · open weights · Free download (GitHub, Hugging Face)The project page lists Tamil (tam_Taml) among the 22 scheduled Indian languages it translates.
Param2-17B-A2.4B-ThinkingBharatGen · open weights · Free download (Hugging Face), non-commercial licenceThe model card lists Tamil among the 21 Indian languages it supports besides English and Hindi.
Sarvam-105BSarvam AI · open weights · Sarvam API, Indus app; free download (Hugging Face, AI Kosh)Sarvam's docs list Tamil (ta-IN) among the 10 Indian languages plus English that Sarvam-105B supports, in native script, romanised and code-mixed input.
Sarvam-TranslateSarvam AI · open weights · Sarvam API; free download (Hugging Face)The model card lists Tamil among the 22 Indian languages it translates to and from English.
Tamil LLaMAAbhinand Balachandran (independent) · open weights · Free download (Hugging Face)The model card describes it as bilingual: English and Tamil.

AI models compared

How Tamil AI is tested

MILU, from AI4Bharat, IIT Madras and IBM Research India, tests knowledge in 11 Indian languages including Tamil using regional and state exam questions. (MILU paper (arXiv))

IndicGenBench, published by Google researchers, tests writing in 29 Indian languages including Tamil, through summarising, translating and answering questions. (IndicGenBench (GitHub))

Vistaar from AI4Bharat is a set of 59 speech recognition tests in 12 languages including Tamil, and AI4Bharat's Indic LLM Arena ranks chatbots by public votes on Indian-language prompts. (AI4Bharat Vistaar (GitHub), AI4Bharat blog)

Tamil in government AI services

Bhashini, the central government's language AI platform, offers voice services in 22 languages and text services in 36. It was used at Kashi Tamil Sangamam to translate Hindi speech into Tamil live. (PIB)

AU-KBC Research Centre built a Tamil-speaking chatbot for TNeGA to help people fill in certificate applications. (AU-KBC Research Centre)

Learn AI in Tamil

IIT Madras offers Introduction to Machine Learning (Tamil) on NPTEL, taught by Prof. Arun Rajkumar in spoken Tamil, with technical terms, tests and assignments in English. Anyone can take it free; a certificate needs a proctored exam. (NPTEL, IndiaAI)

NPTEL translates course material into 11 Indian languages including Tamil, and had translated 174 technical courses into Tamil by April 2024. (NPTEL Translation, DT Next)

Our course finder lists the AI courses taught in Tamil that we track:

Generative AI Course in TamilHCL GUVI · Intermediate · Paid · in TamilRecorded lessons in Tamil on large language models, prompt engineering, LLM APIs and retrieval-augmented generation, ending with a chatbot project. You need working Python.
Introduction to Machine Learning (Tamil)NPTEL (IIT Madras) · Beginner · Free (certificate paid) · in TamilA short first course on the main ideas and algorithms of machine learning, taught in spoken Tamil by Prof. Arun Rajkumar. Technical terms, assignments and exams are in English. Made for Tamil-speaking undergraduates and teachers.
Prompt Engineering for Beginners in TamilHCL GUVI · Beginner · Paid · in TamilA short self-paced course in Tamil on writing good prompts for AI chat tools. Basic knowledge of AI helps.
YUVA AI for ALLIndiaAI Mission (MeitY) · Beginner · Free · in TamilA free national AI literacy course in six modules on what AI is, how to prompt AI tools, and how to use AI safely. Open to anyone who can read and write, including farmers and business owners. Available in Tamil.

All AI courses in Tamil Nadu

Questions people ask

Is there an AI model made for Tamil?

Yes. Tamil-Llama adapted Llama 2 for Tamil in 2023, IIT Madras's AI4Bharat publishes Tamil translation, speech and text models, and Sarvam AI's models support Tamil. The Tamil Nadu government also plans its own Tamil large language model through the Tamil Virtual Academy.

Which AI can translate Tamil to English?

Open options include AI4Bharat's IndicTrans2, which covers all 22 scheduled languages including Tamil. Sarvam-Translate does the same through Sarvam AI's API, and Bhashini, the central government's platform, offers translation in many Indian languages.

Is there Tamil speech-to-text and text-to-speech?

Yes. For speech-to-text there are AI4Bharat's IndicConformer, Sarvam AI's Saaras and Google Cloud Speech-to-Text. For text-to-speech there are AI4Bharat's Indic Parler-TTS, Sarvam AI's Bulbul and Google's Gemini-TTS.

When will the Tamil Nadu government's Tamil LLM be ready?

No date has been given. The revised 2026-27 budget on 5 Aug 2026 announced the plan, led by the Tamil Virtual Academy, without an amount or timeline. We will update this page when the state says more.

Our story on the plan

Can I learn machine learning in Tamil for free?

Yes. IIT Madras offers Introduction to Machine Learning (Tamil) on NPTEL, free for anyone; a certificate needs a proctored exam.

AI courses in Tamil Nadu

Sources

Every fact on this page was checked against these pages, most recently on 22 Oct 2026. Lists marked as live come from our own records, each of which names its source.

Spotted an error? Tell us: we fix confirmed errors within 24 hours and log them in public.

Cite this page

theTN.ai, "Tamil-language AI", updated 22 Oct 2026. https://thetn.ai/tamil-language-ai/