The news
On July 17, 2025 Zoho launched Zia LLM and deepened its AI portfolio with agents, an agent builder, MCP support and an agent marketplace.
Key announcements from the press release include:
In-House LLM: Zoho has developed its own large language model, Zia LLM, which comes in three sizes (1.3B, 2.6B, and 7B parameters) to optimize for different business use cases. This allows customers to leverage AI while keeping their data within Zoho's ecosystem, ensuring privacy. The three models allow Zoho to always optimize the right model for the right user context, striking the proper balance between power and resource management. This focus on right-sizing the model is an ongoing development strategy for Zoho.
Speech-to-Text Models: The company also unveiled two proprietary Automatic Speech Recognition (ASR) models for English and Hindi, with plans to support more languages in the future.
Prebuilt AI Agents: To facilitate immediate adoption, Zoho has introduced a range of AI agents that are integrated directly into its products. These agents are designed to automate tasks for various business roles such as sales development, customer support, and account management.
Global and Private Cloud Deployment: The new Zia LLM will be deployed across Zoho's data centers in the US, India, and Europe.
Continued Support for Other Models: While promoting its own AI, Zoho will continue to support integrations with other popular large language models like ChatGPT, Llama, and DeepSeek.
Zoho will continue to scale Zia LLM’s mode sizes. A2A capabilities are on the roadmap.
The bigger picture
Enterprise software has been a platform game for a long time. AI, in particular generative and agentic AI, have upped the ante in this respect. The competition has become even more one that is based around platforms. An enterprise software vendor that wants to address a significant portion of an enterprise’s value chain cannot go without agents, an agent builder, and a marketplace anymore. This is true whether they use an own AI, a derivative of an open source one, or build it using their own stack.
Plus, support of MCP becomes non-negotiable as most customers run on more than one platform. A2A support then is the next step. Both can also be used as a bridge head and Trojan horse to work towards an increased share of the customers’ value chains.
At the same time this AI battleground is evolving with an enormous pace which seems to accelerate every day. This makes it important for buyers to have a deep look at who to engage with while vendors need to offer a strong narrative around a compelling value proposition.
My point of view and analysis
With this announcement it is clear that Zoho is serious about AI. With this, the company raises its AI profile significantly, particularly in the CX space, as Zoho CRM is its most mature product.
The company works on AI capabilities, tooling and infrastructure tor quite some time now. Plus, it owns its hardware and data centers. This includes already having deployed more than 80 AI algorithms in the past.
Now it also has a family of LLMs – a family that is poised to grow, initially with a 32bn parameter model. And frankly, to me this is the biggest splash although all other parts of the announcement show a round offering.
One of the core questions that I asked myself is whether the world needs another LLM or whether it would be sufficient to adapt an existing open source one to own needs. On the other hand, this would not be the Zoho way. Zoho is adamant in staying in control of its own destiny. As part of this, the company is convinced that it needs to own its software stack, as only this way gives full control, full understanding of its behavior and all options. This is certainly also true for an LLM. As Raju Vegesna, Chief Evangelist of Zoho, said: “If AI is going to be the future, and if you are a technology company, what’s more important than creating that core part of technology that is going to define its future?”
In addition, Zoho’s LLM family comes with a twist. It is not addressing consumer tasks but is specialized on what Zoho customers require. And Zoho customers are predominantly B2B businesses. In that sense, and true to the objective of delivering “contextual AI”, Zoho follows a route that is similar to the one that SAP takes with its own foundation model. In contrast to SAP, Zoho did not train the model on customer data (SAP trains its foundation model with customer consents) which is in line with the company’s strong stance on privacy and trust. Offering differently sized models is all about “rightsizing”, i.e., the economic and performant use of valuable compute resources. In essence, Zoho offers to not bring a sledgehammer to a job when a mallet is sufficient. If the system is able to select the right model automatically, this can be a game changer.
The second really interesting part of the announcement is about its automatic speech recognition (ASR) models. This is again a matter of being independent. Further, a well-functioning ASR can help businesses raise significant efficiencies, if the models work well. As per Zoho’s tests they compare well against larger models. The current limitation is that the ASR models support only English and Hindi at the moment. I expect further widely used languages being supported soon.
The offering of prebuilt agents, an MCP server, an agent studio and a marketplace show that Zoho is serious, but they are also table stakes these days. They are the foundation for scale. These offerings, in combination with the upcoming A2A support, open up the ecosystem game by enabling partners to innovate fast and to facilitate getting into new accounts by being open.
All in all, this is a bold, yet not unexpected move by Zoho. It helps the company to strengthen its enterprise credibility, while maintaining the needs of SMBs squarely in the sights. Especially considering Zoho’s strong stance on value and pricing, the company now places significant pressure on the competition but also on itself, as it is still testing whether it can deliver these capabilities as part of the subscription or whether at least parts of it need to get priced. But just doing this exercise with the goal of delivering these capabilities as part of the package. Even if Zoho will not be able to absorb the cost of running models and agents, pricing can be expected to be extremely competitive.
As Raju Vegesna said, it is a marathon.
And this marathon has only just started.
Comments
Post a Comment