Archive · 2 Oct 2026
Sarvam Unveils Speech AI for India’s 22 Languages
Sarvam is an Indian company building artificial intelligence for India.
It released a speech tool called Saaras V4.
The company says the tool understands all 22 of India’s scheduled languages and English.
This could help people use computers by speaking in their own languages.
It may be especially useful for people who find English or typing difficult.
Sarvam is also working with IBM on AI systems for government and regulated businesses.
These systems are meant to keep important data and computer systems under India’s control.
However, supporting a language does not necessarily mean the tool works equally well in every language.
The partnership has also been announced but is not yet proof of nationwide government use.
Indian startup Sarvam says its Saaras V4 speech-recognition model supports all 22 scheduled Indian languages plus English.
The model is intended to improve voice-based AI access for people who do not primarily use English or face barriers typing in local scripts.
Sarvam is partnering with IBM to develop a sovereign AI stack for Indian government departments and regulated industries.
The partnership aims to keep data and systems under Indian legal and physical control rather than relying entirely on foreign cloud infrastructure.
The language coverage and sovereignty claims still require independent testing and large-scale deployment to assess quality and practical impact.
- Who
- Indian AI startup Sarvam, in partnership with IBM.
- What
- Sarvam released Saaras V4, a speech-recognition model it says supports India’s 22 scheduled languages and English, and announced a sovereign AI partnership with IBM.
- Where
- India, with the proposed systems intended for Indian government departments and regulated industries.
- Why
- To expand AI access across India’s languages and help sensitive organizations use AI under Indian legal and physical control.
Potential benefits
Caveats and limitations
Language access
Potential benefits
Supporting all 22 scheduled languages could make voice-based AI available to a much larger share of India’s population, including people who do not use English or face typing barriers.
Caveats and limitations
The company’s coverage claim does not show that performance is equally strong across all 22 languages; independent testing is still needed.
Sovereign AI
Potential benefits
A domestic AI stack could allow government and regulated organizations to use AI while keeping data and systems under Indian legal and physical control.
Caveats and limitations
The stack still relies, for now, on some hardware and software originating abroad, so its independence is partial.
Government impact
Potential benefits
The IBM-Sarvam partnership could provide infrastructure for sensitive uses such as administration, welfare and records.
Caveats and limitations
An announced partnership is not yet evidence that the system has been deployed across government at scale.
- Company
- Sarvam
- Speech model
- Saaras V4
- Claimed language coverage
- All 22 of India’s scheduled languages, plus English
- Partner
- IBM
- Intended users
- Indian government departments and regulated industries
- Sovereignty objective
- Keeping data and systems under Indian legal and physical control
- Key caveat
- Coverage does not establish equal quality across all languages, and the partnership is not yet a large-scale deployment