Return to page

H2O Danube, a super-tiny, open-weights small language model for on-device and offline applications surpasses Apple and rivals Microsoft with latest Danube3-4B release, achieving over 80% accuracy on 10-shot HellaSwag benchmark

The newly released H2O-Danube3 is now available worldwide on Hugging Face. This latest addition to the H2O SLM series includes both the H2O-Danube3-4B and the H2O-Danube3-500M models.

These models aim to democratize advanced NLP capabilities and showcase remarkable versatility and efficiency, making them ideal for chatbots, research, and on-device applications.

H2O-Danube3-4B is Available Worldwide Learn MoreGet the Mobile App

Phone screen showing open source Danube3-4B next to a chart comparing academic benchmarks from H2O vs Alibaba,, Apple, and Microsoft.

Convergence of the world’s best predictive and generative AI

Supercharge human productivity

Enterprise h2oGPTe synthesizes diverse data and lets you process thousands of documents or web pages–any way you want: inspect, extract, translate, transform, find differences, summarize. Your imagination is the limit.


Try Enterprise h2oGPTe Freemium Learn More

Multi-modal Document AI with enterprise scale and security

Convert structured data like financial tables to JSON

Get answers from diagrams, flowcharts, web pages, audio files, video files and images.

web of yellow data lines with a lock and shield web-of-yellow-data-lines-with-a-lock-and-shield

Built for air-gapped, on-premises, private or managed cloud deployments.

Own your data, own your prompts.

Only provides an end-to-end GenAI platform where you can own every part of the stack. offers a highly flexible solution, that can run fully managed in the cloud, or in hybrid or on-premise and air-gapped environments.

Fully scalable with Kubernetes.

AICPA SOC2 compliance seal AICPA SOC2 compliance seal achieves 

Cost control and flexibility

Customers can decide to mix and match from a large choice of over 30 LLMs:

Proprietary LLMs:

  • Gemini, Claude, GPT-3.5 or 4, Bedrock, Groq, Mistral, etc.

  • Reliance on third-party provider

  • No control, no ownership

  • Token-based pricing, unbounded cost

Open source LLMs:

  • Mixtral, Mistral, H2O Danube, Llama3, Fine-Tunes, etc.

  • Any model from Open-Source ecosystem

  • Hosted on your own GPU infrastructure

  • Full control and ownership of models

  • Fixed hardware cost, most economical choice

  • Cost savings on Platform. 25x cheaper on query cost.

  • Serve more users. Better accuracy with up to 100% cheaper on Document Processing.

product shot of a drop down menu under "LLM to use" with a list of models to choose from product shot of a drop down menu under "LLM to use" with a list of models to choose from
white arrow pointing up right white arrow pointing up right
h2oGPTe is designed to work with any LLM of your choice
gartner logo in white gartner logo in white

Visionary in Gartner Magic Quadrant: 
Data Science and Machine Learning 2024

Listed in Gartner Magic Quadrant (Cloud AI Developer Services) in 2020, 2021, 2022, 2023, and 2024

Listed in Gartner Magic Quadrant (Data Science and Machine Learning) in 2018, 2019, 2020, 2021, 2023, and 2024




Forrester logo white Forrester logo white

The Forrester Wave™: Computer Vision Tools, Q1 2024:

Named as a leader in this year's Forrester’s Wave report for the Computer Vision Tools Landscape







GenAI App Store

As the maker of the world’s best AutoML, is converging Predictive AI with Generative AI, and helping leading global brands save billions of dollars a year with purpose-built AI for hundreds of use cases such as:

Fraud and scam detection

Complaint summarizer

Recommendation engine

KYC, loan application risk

Pricing engines and more


screenshot of code, black text on white screenshot of code, black text on white

Get your API key

H2O Generative AI can solve a broad spectrum of natural language use cases, including classification, semantic search, paraphrasing, summarization, and content generation.

h2oGPTe is used by companies to build production-ready, scalable and efficient AI-powered applications.


Write custom apps with fully featured Python client APIs 

Seamlessly integrate state-of-the-art text, image, and code generation capabilities into your projects