New Delhi, 29 August (APAC Media): Voice AI startup Gnani AI has launched an open-weight foundational large language model (LLM), Evon v3.3, along with a new agentic AI platform, Plexus, as part of its end-to-end sovereign AI stack, Artha.
The platforms were unveiled in New Delhi in the presence of Vice-President C. P. Radhakrishnan at his official residence. Speaking at the event, Radhakrishnan described sovereign LLMs as a major advantage for India and stressed the need to make AI open, affordable, and accessible.
Evon v3.3 Built for Indian Languages
Evon v3.3 is a 30-billion-parameter model, with around 3.5 billion parameters active for any given token. This indicates a mixture-of-experts (MoE) architecture designed to improve efficiency while supporting reasoning and other AI tasks.
The model supports more than 11 Indian languages, with its weights released under the Apache 2.0 licence. Developers can request the weights for free through Hugging Face.
A key focus of Evon v3.3 is token efficiency. According to Gnani AI, the model uses around 20% fewer tokens per Indian-language word than the tokeniser used by the GPT-5 family and less than half the tokens required by byte-level tokenisers used by models such as DeepSeek, Llama, and Qwen.
Gnani said it achieved this by developing a new tokeniser specifically optimised for Indian scripts, potentially reducing AI inference costs for enterprises.
Plexus Enables AI Agent Deployment
Alongside Evon v3.3, Gnani launched Plexus, an enterprise-focused agentic AI platform. It allows businesses to build and deploy AI agents using natural-language prompts.
The agents can perform tool calling and operate autonomously across documents, enterprise systems, and conversations. Plexus also allows customers to select from multiple underlying models, including Evon v3.3.
Together, the two offerings from Gnani will provide Indian organisations with a locally focused and cost-efficient sovereign AI stack.
Also Read
TCS, Porsche Forge Partnership to Accelerate AI-Powered Mobility
