In a significant move to strengthen its AI capabilities, Zoho Corporation has unveiled Zia LLM, a proprietary large language model developed completely in-house. Announced on July 17, 2025, this launch represents Zoho's strategic investment of approximately $20 million in building foundational AI technology that integrates seamlessly with its extensive software ecosystem.
Zia LLM comes in three distinct parameter sizes—1.3B, 2.6B, and 7B—each separately trained and optimized for different business use cases rather than being distilled versions of a single model. Built entirely in-house using a mix of public and proprietary datasets with a GPT-3-style architecture, the models were trained on datasets ranging from 2T to 4T tokens. The LLM is tailored specifically for business applications such as structured data extraction, summarization, Retrieval Augmented Generation (RAG), and code generation, with each model optimized for different levels of contextual complexity.
Alongside Zia LLM, Zoho introduced Zia Agents, with over 25 ready-to-deploy AI-powered agents available in Agent Marketplace; Zia Agent Studio, a no-code agent builder; and a model context protocol (MCP) server to open up Zoho's vast library of actions to third-party agents. These capabilities are designed to help organizations fully realize and maximize the value of contextual, assistive, and agentic AI technology, immediately impacting daily workflows for diverse roles and use cases.
The Zia Agent Studio experience has been simplified to be fully prompt-based (with the option to use low-code) and includes ready-made access to over 700 actions across Zoho's products. Agents built by users can be deployed autonomously, triggered through button clicks or rule-based automation, or summoned within customer conversations. At deployment, an agent can be provisioned as a digital employee that respects defined user access permissions. Administrators can perform behavioral audits and analyze performance to ensure agents operate effectively within clear guardrails.
Zoho's approach emphasizes data privacy, with Zia LLM allowing customers to keep their data on Zoho servers rather than sending it to external AI cloud providers. The model will be deployed across Zoho's data centers in the US, India, and Europe, and is currently being tested for internal use cases across Zoho's broad app portfolio, with customer availability expected in the coming months.
Looking ahead, Zoho plans to regularly scale Zia LLM's model sizes, with the first of several planned parameter increases by the end of 2025. Future releases will include expanding the available languages used by the speech-to-text model, introducing a reasoning language model (RLM), adding skills to Ask Zia for finance and customer support teams, and implementing Agent2Agent (A2A) protocol to allow Zia Agents to interact with each other and with agents on other platforms.