Gnani ArthaThe next frontier of Sovereign AI

Launched by the Hon’ble Vice-President of India

Gnani Artha was unveiled by Shri C. P. Radhakrishnan, Hon’ble Vice-President of India, in New Delhi on 28 August 2026.

Bharat is making significant strides in AI. I congratulate Gnani AI on this significant milestone and I am pleased to launch Gnani Artha, which brings together the power of Evon, a large language model, and Plexus, a platform that connects this intelligence with real-world work and institutions. Both these platforms reflect the growing strength of India’s technology ecosystem. May this initiative continue to building a stronger, more self-reliant, and technologically empowered India.

Shri C. P. RadhakrishnanHon’ble Vice-President of India

The Sovereign AI Mission: Solving India’s High-Stakes Data, Language, and Operational Complexities

An application filled in Marathi. Six months of bank statements. GST filings. Photographs of several identity documents. Reading it takes judgement, not transcription.

Cross-checking declared income against actual turnover. Finding the inconsistencies that matter. Every lender in India handles thousands of these a day, and almost none can do it with AI that reasons well enough in the applicant’s own language, at a cost sustainable at that scale, and without the file ever leaving their own systems.

Intelligence that reasons in one’s own language, economics that survive thousands of files a day, and a way to put both into production inside the lender’s own systems. Indian institutions have had these in fragments.

Gnani Artha brings them together as one stack.

Sovereign Intelligence

Reasons in Indian languages

Not English-first with translation bolted on. Trained natively across eleven Indian languages, on Indic data.

Frontier Efficiency

Survives Indian volumes

Indian scripts cost more to process than English. Artha attacks that language tax at the tokenizer — every model benchmarked against Evon v3.3 costs at least 2.2× more per point of Indian-language accuracy.

Complete Control

Runs inside your walls

Open weights on a single node. DPDP, RBI and IRDAI residency met by where the model runs.

One stack, under your control

Artha is Gnani AI’s end-to-end sovereign AI solution — the models, the platform, and the infrastructure they run on.

The stack Gnani Artha Sovereign frontier-grade intelligence, AI economics that work in India, and real-world impact at scale.
Models Four sovereign models Built in India, trained on Indic data, deployable on your own infrastructure.

Gnani Evon v3.3

Language & reasoning

30B open-weights model across 11 Indian languages. Cuts the language tax on Indian AI.

Gnani Evon v2.0

Enterprise reasoning

Unified reasoning and tool orchestration across 11+ Indic languages, calibrated for BFSI, insurance, healthcare and telecom.

Gnani Prisma v2.5

Speech recognition

Indic ASR trained on real telephonic audio, built for code-switching and regional accents.

Gnani Timbre v2.5

Speech synthesis

Neural synthesis with context-aware prosody, streaming-ready across Indian languages.

Platform Gnani Plexus The agentic AI workforce. Turns that intelligence into governed, auditable outcomes.
Deployment Your own infrastructure Inside your data centre or VPC. Customer data never leaves your network.

Intelligence

Gnani Evon v3.3

The sovereign LLM that cuts the language tax on Indian AI.

On MILU — a widely used Indian-language benchmark spanning eleven languages and dozens of academic and professional subjects — Evon v3.3 outperforms a 105-billion-parameter Indic model on ten of eleven languages, and a similarly sized 30B model on all eleven, achieving parity with a similar-size hosted global frontier model.

To make the economics work, Gnani AI rebuilt the model’s tokenizer — the layer that breaks language into the units a model processes — for Indian scripts. Fewer tokens mean lower compute cost, faster responses, and more usable context per query.

11

Indian languages, trained natively

30B / 3.5B active

Total parameters, and what you pay to serve

~20% fewer

Tokens per Indian word vs the GPT-5 family tokenizer

1 node

Self-hosted, open weights, Apache 2.0

Weights
Open, Apache 2.0
Availability
By request on Hugging Face

Agentic workforce

Gnani Plexus

Turning sovereign intelligence into real-world impact.

Each agent on Plexus is a discrete, identity-bearing unit — closer to an employee than a script — combined with other agents into workflows built around a defined outcome. Those workflows run under an orchestration layer that can be human-in-the-loop or AI-led.

  • Agentic AI workforce

    Built for your needs.

  • Trigger-based workflows

    Run in parallel or sequentially.

  • Multi-system integration and tool-calling

    Plug in the systems you already run.

  • Enterprise-grade AI governance

    Prompt defense. Jailbreak resistance. Data protection. Output control.

Government and Public Sector Use Cases

Grievance resolution & pattern detection

A multilingual agent captures a citizen’s complaint in their own language. A reasoning agent detects patterns across recent complaints — a spike in one district, one recurring failure — files a single consolidated ticket with the department that owns it, and closes the loop with the citizen once resolved.

Beneficiary scheme enrolment & outreach

Plexus cross-references a welfare scheme’s eligibility criteria against population and beneficiary databases to identify citizens who qualify but haven’t enrolled — then reaches out in the citizen’s own language, explaining how to complete enrolment, or where an application is stuck and what to do next.

Subsidy & benefit transfer failure resolution

When a direct benefit transfer fails — a pension, a scholarship, an LPG subsidy — usually on a bank or Aadhaar mismatch, Plexus flags it automatically. An agent calls the citizen in their own language, walks them through the fix, and Plexus rechecks the transfer status until it clears.

Disaster & emergency response triage

Following a flood or disaster alert, outbound calls run at scale across the affected area, capturing what each household needs — shelter, food, medical help — in their own language. Plexus triages by severity and routes each case to the right relief agency, rather than a single generic helpline queue.

Public health outreach & follow-up

Vaccination, TB and maternal-health reminder calls with basic symptom triage, in-language. Plexus escalates any red-flag response to the nearest health centre and assigns a follow-up task to the local health worker automatically.

Government financial reconciliation

Plexus matches scheme disbursement records across treasury systems, bank confirmations and payment logs, clears the clean matches, and drafts a root-cause note on each genuine discrepancy for the audit or finance team to review.

Enterprise Use Cases

Document-driven underwriting

Reads loan applications, bank statements, GST filings and identity documents, cross-validates them against bureau and Account Aggregator data, and approves, declines or refers each file within governed thresholds.

Reconciliation, banking

An agent matches millions of line items across bank statements, core-banking ledgers and payment-switch logs, drafts a root-cause narrative for each genuine exception, and routes it to the team that owns it — automating one of the highest-cost manual functions in Indian banking.

KYC & compliance remediation

Scans KYC repositories for expiring or missing fields, triggers re-KYC workflows automatically, and closes the loop without anyone having to initiate it.

Dispute & chargeback resolution

Ingests dispute claims and their supporting evidence, applies card-network rules, resolves straightforward cases autonomously, and routes contested ones to a human.

Voice-authenticated instant disbursal

Voice biometrics verifies identity, the agent confirms live eligibility and captures verbal consent, and disbursal is triggered on the call — collapsing a loan or top-up into a single conversation, with no branch visit or app step.

Grievance intake & pattern escalation

Multilingual inbound captures the complaint; a reasoning agent detects patterns across calls — a spike in failed UPI reversals, say — and files one bulk ticket instead of many, then calls back to close the loop.

Illustrative of platform capability. None of these describes a specific existing deployment, customer or government department, and none implies partnership with or endorsement by any government body.

Deployed inside your own infrastructure

Gnani Artha runs where your data already lives — your data centre or your virtual private cloud.

Single node

Roughly 3.5 billion parameters active on any given token, so Evon v3.3 fits on one inference node rather than a cluster.

Residency by architecture

DPDP, RBI and IRDAI residency requirements are met by where the model runs — not by a clause in a contract.

Governed and auditable

Guardrails, observability and audit logging built in from the first deployment rather than added afterwards.