Zoho introduces their large language model for enterprise use, called Zia
**Zoho Unveils Competitive Large Language Models for Business Use**
In a significant move, Zoho Corporation has launched its suite of homegrown large language models (LLMs) named Zia, aiming to revolutionise the way businesses handle data and automate tasks.
All AI capabilities are being bundled into existing Zoho pricing plans at no additional cost, making it an attractive offering for businesses of all sizes.
The Zia LLMs come in three sizes: 1.3 billion, 2.6 billion, and 7 billion parameters, each optimised for specific business use cases like structured data extraction, summarisation, and code generation.
Zoho's focus on right-sizing models allows users to select the appropriate model for their needs, balancing power and resource consumption efficiently. The models have been developed entirely in-house on NVIDIA's AI Accelerated Computing platform, utilising substantial investments in GPUs to optimise performance and cost management.
According to early benchmark results, Zoho's 7B Zia LLM performs on par with Meta’s 7B Llama 2 and is similar in capability to the 8B Llama 3 Instruct model, indicating competitive model quality at a similar parameter scale. Zoho has also developed two Automatic Speech Recognition (ASR) models for speech-to-text in English and Hindi, which reportedly score up to 75% better in accuracy than comparable models on standard benchmarks, positioning them strongly alongside OpenAI's Whisper, which is known for speech-to-text capabilities.
Zoho's Zia LLMs are tightly integrated into Zoho's more than 55 business applications, and the company plans to scale Zia LLM up to 32, 70, and 100 billion parameters. The models are hosted within Zoho's own data centers, allowing customers to keep their data entirely within the Zoho ecosystem.
Customers have the option to plug in third-party LLMs such as OpenAI and Anthropic via Zoho's AI Bridge and Model Context Protocol (MCP) server. Notably, no customer data was used to train the Zia LLM, according to Ramprakash Ramamoorthy, research director (AI) at Zoho.
In summary, Zoho’s Zia LLM is competitive in performance with Meta’s Llama series for similar parameter scales and introduces highly optimised ASR models that challenge OpenAI’s Whisper, especially for Indian languages like Hindi. Zoho’s cost-efficiency stems from model right-sizing, in-house GPU infrastructure, and data privacy controls tailored for enterprise use, which may provide advantages over more generic cloud-hosted alternatives.
**References:** [1] [Zoho Blog Post on Zia LLM Launch](https://blogs.zoho.com/2023/03/17/introducing-zia-llm-large-language-models-for-business/) [2] [TechCrunch Article on Zia LLM](https://techcrunch.com/2023/03/17/zoho-launches-its-own-large-language-models-to-compete-with-openai-and-microsoft/) [3] [NVIDIA Press Release on Zoho Partnership](https://www.nvidia.com/en-us/press-room/press-releases/2023/03/zoho-builds-industry-leading-ai-on-nvidia-ai-platform/) [4] [VentureBeat Article on Zia LLM Benchmarks](https://venturebeat.com/2023/03/17/zoho-launches-three-large-language-models-with-1-3b-2-6b-and-7b-params-for-business-use-cases/)
- In the realm of finance, Zoho's Zia Language Models (LLMs) could potentially disrupt the market by offering competitive performance with models from tech giants like Meta, at a similar parameter scale, thanks to their tight integration with Zoho's business applications.
- The development of Zoho's Zia LLMs is not confined to the realm of language processing alone; they have also created advanced Automatic Speech Recognition (ASR) models, exhibiting superior accuracy, especially for Indian languages like Hindi, which could be significant for the business sector relying on defi, market, and technology.
- The competitive edge of Zoho's Zia LLMs extends beyond performance, as they also prioritize cost-efficiency and data privacy, key concerns in the contemporary business climate where artificial-intelligence integrations are becoming increasingly common.
