Models
AI Model Hub for Free: From December 1, 2024, to September 30, 2025, IONOS is offering all foundation models in the AI Model Hub for free. Create your contract today and kickstart your AI journey!
We continually strive to bring you the best and latest AI technologies, ensuring you have access to cutting-edge solutions that deliver the optimal experience. Therefore, we continuously update our model catalog by introducing the latest models and retiring older ones.
This page provides a detailed list of models, categorized by their current status, to keep you informed about upcoming changes in model availability.
Active Models
The following models are currently active in the AI Model Hub with no scheduled retirement date:
Large Language Models
Small models (less than 15 billion parameters)
The following small language models (less than 15 billion parameters) are optimized for fast inference and low resource consumption. They are ideal for real-time applications and scenarios where latency and cost are critical. While their compact size means they may have less "world knowledge" and slightly lower response quality than larger models, they excel at conversational tasks, virtual assistance, and domain adaptation.
openGPT-x Teuken: A highly adaptable and lightweight model well-suited for conversational agents and virtual assistants across diverse domains. Its efficient architecture enables quick responses and easy customization, making it a strong choice for research and production environments where flexibility is key.
Meta Llama 3.1 8B: A US-developed model by Meta, designed for conversational agents and virtual assistants. It balances speed and language understanding well, making it suitable for interactive applications that require reliable, fast responses.
Mistral Nemo: A French model from Mistral, tailored for conversational agents and virtual assistants. Its small size ensures efficient inference, and it is particularly effective in scenarios where rapid, context-aware dialogue is essential.
Medium models (between 15 billion and 150 billion parameters)
Medium-sized models provide a strong balance between response quality and inference speed. They are suitable for applications that demand higher accuracy, broader knowledge, and more nuanced language understanding while maintaining reasonable performance and cost.
Meta Llama 3.3 70B: A US model by Meta, offering enhanced response quality for conversational agents and virtual assistants. Its larger parameter count enables more sophisticated reasoning and richer language capabilities, making it ideal for demanding dialogue systems.
Mixtral 8x7B: A legacy mixture-of-experts model from Mistral, designed for high-quality conversational and virtual assistant tasks. Its architecture improves response diversity and accuracy, especially in complex or specialized domains.
Large models (more than 150 billion parameters)
Large models are designed for maximum language understanding, deep reasoning, and high-quality responses. They are best suited for advanced applications where accuracy and depth of knowledge are paramount, though they require more computational resources and have slower inference speeds.
Meta Llama 3.1 405B: Meta's flagship large language model, delivering exceptional response quality and comprehensive knowledge coverage. It is ideal for research, content generation, and complex conversational tasks where the highest level of language proficiency is required, albeit with slower response times.
Coding models
Coding models are specialized for code generation, completion, and understanding. They are optimized to assist developers with programming tasks, offering context-aware suggestions and code snippets.
Code Llama 13B: A medium-sized coding model from Meta, tailored for code generation and completion. It provides reliable support for various programming languages and tasks, balancing performance and accuracy for developer productivity.
Embedding Models
Embedding models convert text into dense vector representations, enabling semantic search, clustering, and similarity comparison. They are essential for building search engines, recommendation systems, and knowledge retrieval applications.
BGE Large 1.5: Maps English text to a 1024-dimensional dense vector space, supporting high-quality semantic search and retrieval tasks.
BGE m3: Handles long, multilingual text, mapping it to a 1024-dimensional vector space for cross-lingual search and analysis.
Paraphrase Multilingual MPNet v2: A sentence transformer model that maps text to a 768-dimensional vector space, enabling efficient paraphrase detection and semantic similarity tasks.
Text To Image Models
Text-to-image models generate images from textual descriptions, supporting creative and generative AI use cases such as content creation, design, and prototyping.
FLUX.1-schnell: A German model by BlackForestLabs, optimized for fast and high-quality image generation from text prompts, suitable for creative and commercial applications.
Stable Diffusion XL: A US model from stability.ai, renowned for generating detailed and diverse images from text, widely used in digital art and design workflows.
Scheduled for Retirement
The following models are scheduled to be retired. After the specified date, they will no longer be available to use in the AI Model Hub:
Model Provider
Model Name
Scheduled to Retire On
Alternative
To avoid disruption and ensure long-term support, please replace models scheduled for retirement in current and new projects with suitable alternatives.
Retired Models
These models are no longer available to use in the AI Model Hub:
Model Provider
Model Name
Retired On
Alternative
Last updated
Was this helpful?