H2O.ai Multimodal Models Surpass One Million Monthly Downloads

KUALA LUMPUR, March 17 (Bernama) -- H2O.ai, a pioneer in sovereign artificial intelligence (AI), announced its openly available H2OVL Mississippi visual-language multimodal models have reached a significant milestone, surpassing one million monthly downloads.

Hosted on Hugging Face, the H2OVL Mississippi models have rapidly emerged as one of the most adopted open multimodal OCR‑capable models in the ecosystem, matching leading category performers such as Qwen, DeepSeek, and Google’s Gemma.

This surge in adoption reflects a growing shift towards sovereign, open, transparent and locally deployable AI models, a direction H2O.ai has championed for more than a decade, according to a statement.

bootstrap slideshow

Key features of H2OVL Mississippi 2B and 0.8B include parameters that are optimised for efficient deployment, enabling strong performance with minimal resource consumption and cost.

The models also handle OCR and Document AI tasks across varied resolutions, providing versatile vision-language capabilities. They incorporate multi-stage training with fine-tuning layers to support highly customised application performance.

In addition, the models deliver real-time processing with minimal latency, making them suitable for industries such as banking, financial services, telecommunications, manufacturing and healthcare, where accurate document processing is critical.

bootstrap slideshow

According to the FutureX February 2026 Overall Leaderboard, H2O.ai leads the field with a score of 59.1 per cent, outperforming major AI developers and positioning itself as a top-ranked AI model ecosystem.

Furthermore, the company has been recognised by Gartner as a Visionary in the 2025 Gartner Magic Quadrant for Data Science and Machine Learning Platforms for the third consecutive year, citing its completeness of vision and ability to execute.

-- BERNAMA