Zoho Corp is set to unveil two artificial intelligence-based foundational models by the end of the year, according to group CEO Shailesh Davey. These models will feature 7 billion and 13 billion parameters, respectively. Davey, who assumed the role in January after co-founder Sridhar Vembu transitioned to chief scientist for AI projects, emphasized that these foundational models are integral to the company’s product offerings and will be publicly accessible within the next two quarters.
In addition to these models, Zoho is also developing foundational models tailored for Indian languages, although a launch timeline has not yet been established. Davey noted that the foundational models serve as technology demonstrators, showcasing the company’s confidence in its capabilities.
One of the challenges in creating foundational models is the increased demand for computational power and the necessity for high-quality training data. Zoho is focusing on optimizing these models to enhance power efficiency. Unlike consumer-focused models, business-to-business models are designed to meet specific industry needs and may not require as many parameters. For comparison, ChatGPT-4 has 175 billion parameters, while the anticipated ChatGPT-5 is expected to reach 1.5 trillion parameters. The Chinese model DeepSeek R-1 was released with 671 billion parameters, alongside smaller versions ranging from 1.5 billion to 70 billion parameters.
Currently, Zoho provides access to various models, including DeepSeek and OpenAI, with capabilities of up to 32 billion parameters through its platform. Customers can submit queries, and the platform selects the appropriate model for responses. The two new foundational models will also be integrated into the Zoho platform later this year.
Davey highlighted the benefits of smaller models, noting that they require less GPU power and can maintain accuracy while being more efficient.
