AI Models: the multi-model platform with ready-to-use AI APIs

Integrate AI into products and workflows with no fixed infrastructure costs and in compliance with the AI Act, within the Aruba Cloud ecosystem.

  • APIs compatible with OpenAI standards
  • Full control over data in our data centers in Italy
  • Consumption model (pay-per-token)
EXPLORE MODELS Get in touch

Bring AI into your applications with less complexity and more control

AI Models gives you access through simple API endpoints: no complex configurations to manage, no upfront hardware investment, scalable resources based on your needs, and the confidence that your data remains protected in our proprietary data centers.

Ready-to-use APIs for AI integration

Connect your applications to the available endpoints and integrate AI capabilities into software, portals, chatbots, workflows or internal tools.

Immediate scalability

Quickly launch PoCs, prototypes and experiments. Move your project from testing to production with scalable AI APIs that support progressively increasing volumes and requests.

Endpoints for different use cases

Generate text, build conversational assistants, analyse and classify content, implement semantic search and much more with AI models ready for your use cases.

Pay-per-token model with no fixed costs

Use AI API endpoints with no fixed infrastructure costs and pay based on consumption. Start with tests or a PoC and scale the service according to your actual usage.

Models and pricing

Choose your AI API endpoints

bge-m3
LIGHT

EMBEDDINGS

bge-m3

Multilingual, multifunctional and multigranular embedding model developed by BAAI. Supports dense, multi-vector and sparse retrieval on texts of varying length in more than 100 languages.

€ 0.0082 + VAT / Mtoken (input)

Parameters: ~567M

Max context: 8.192 token

License: MIT

Features: Dense retrieval, Sparse retrieval, Multi-vector (ColBERT)

Deepseek R1 32B
MEDIUM

REASONING LLM

Deepseek R1 32B

32-billion-parameter distilled model based on Qwen2.5, trained with reasoning data generated by DeepSeek R1. Performs strongly on mathematics, code and reasoning benchmarks.

€ 0.2374 + VAT / Mtoken (input)

€ 0.2374 + VAT / Mtoken (output)

Parameters: 32B

Max context: 128K token

License: MIT

Features: Reasoning, Chain of thought

Gemma-4-31B-IT-NVFP4
MEDIUM

VISUAL LLM

Gemma-4-31B-IT-NVFP4

31-billion-parameter dense multimodal model optimized by NVIDIA in NVFP4 format for efficient inference. Supports text, images and video, advanced reasoning, coding and function calling with context up to 256K tokens.

€ 0.0982 + VAT / Mtoken (input)

€ 0.2948 + VAT / Mtoken (output)

Parameters: 30,7B

Max context: 256K token

License: Apache 2.0

Features: Function calling, Multimodal (images + video), Reasoning, Code assistant

gpt-oss-120b
LARGE

REASONING LLM

gpt-oss-120b

Advanced model for complex reasoning, accurate instruction following and agentic capabilities. Delivers high-level performance for both general-purpose and specialized applications.

€ 0.0319 + VAT / Mtoken (input)

€ 0.1474 + VAT / Mtoken (output)

Parameters: 117B (5,1B active)

Max context: 131K token

License: Apache 2.0

Features: Function calling, Reasoning

intfloat/multilingual-e5-large
LIGHT

EMBEDDINGS

intfloat/multilingual-e5-large

Multilingual text embedding model based on XLM‑RoBERTa‑large, with 24 layers and a 1024-dimensional embedding. Supports 100 languages for retrieval, semantic similarity and classification.

€ 0.0082 + VAT / Mtoken (input)

Parameters: ~560M

Max context: 512 token

License: MIT

Features: Embedding retrieval, Semantic similarity

Llama-3.1-8B-Instruct
LIGHT

LLM

Llama-3.1-8B-Instruct

8-billion-parameter instruction-tuned model optimized for multilingual dialogue. Part of the Llama 3.1 family, it outperforms many open-source models in the same parameter range.

€ 0.0164 + VAT / Mtoken (input)

€ 0.0409 + VAT / Mtoken (output)

Parameters: 8B

Max context: 128K token

License: Llama 3.1 Community License

Features: Function calling, Multilingual

Mistral-Nemo-Instruct-2407
LIGHT

LLM

Mistral-Nemo-Instruct-2407

12-billion-parameter instruction-tuned LLM developed with NVIDIA. Designed for multilingual applications, with strong conversational, code generation and complex instruction-following capabilities.

€ 0.0164 + VAT / Mtoken (input)

€ 0.0246 + VAT / Mtoken (output)

Parameters: 12,2B

Max context: 128K token

License: Apache 2.0

Features: Function calling

Mistral-Small-3.2-24B-Instruct-2506
MEDIUM

VISION LLM

Mistral-Small-3.2-24B-Instruct-2506

24-billion-parameter model with advanced visual understanding and a context window up to 128K tokens. Strong performance on both text and vision tasks.

€ 0.0614 + VAT / Mtoken (input)

€ 0.1637 + VAT / Mtoken (output)

Parameters: 24B

Max context: 128K token

License: Apache 2.0

Features: Function calling, Multimodal

Qwen 2.5 7B instruct
LIGHT

VISUAL LLM

Qwen 2.5 7B instruct

7-billion-parameter vision-language multimodal model that analyzes text, charts, icons and layouts in images. Supports long-video understanding and visual localization with structured output.

€ 0.0327 + VAT / Mtoken (input)

€ 0.0819 + VAT / Mtoken (output)

Parameters: 7B

Max context: 32K token, extendable with YaRN

License: Apache 2.0

Features: Multimodal (images + video), OCR

Qwen3.5-122B-A10B-FP8
LARGE

VISUAL LLM

Qwen3.5-122B-A10B-FP8

Mixture-of-Experts vision-language multimodal model with 122 billion total parameters and 10 billion active. Designed for native multimodal agentic applications across text, images and video.

€ 0.2128 + VAT / Mtoken (input)

€ 1.7027 + VAT / Mtoken (output)

Parameters: 122B (10B active)

Max context: 262K token

License: Apache 2.0

Features: Function calling, Multimodal (images + video), Reasoning, Agents

Qwen3-Coder-30B-A3B-Instruct
MEDIUM

CODE LLM

Qwen3-Coder-30B-A3B-Instruct

LLM specialized in software development with advanced agentic capabilities. Supports long contexts for large repositories and strong performance in code generation, review and automation.

€ 0.057 + VAT / Mtoken (input)

€ 0.209 + VAT / Mtoken (output)

Parameters: 30B (3B active)

Max context: 256K token

License: Apache 2.0

Features: Function calling, Code assistant

Qwen3-Coder-Next
LARGE

CODE LLM

Qwen3-Coder-Next

80-billion-parameter model with only 3 billion active, built for coding agents and local development. Excels at tool calling, adapting to scaffolds and templates, and recovering from errors.

€ 0.057 + VAT / Mtoken (input)

€ 0.209 + VAT / Mtoken (output)

Parameters: 80B (3B active)

Max context: 256K token

License: Apache 2.0

Features: Function calling, Code assistant

Qwen3Guard-Gen-8B
LIGHT

LLM GUARD

Qwen3Guard-Gen-8B

8-billion-parameter content moderation model based on Qwen3. Classifies prompts and responses into three safety levels—safe, controversial and unsafe—with support for more than 119 languages.

Free

Free

Parameters: 8B

Max context: 32K token

License: Apache 2.0

Features: Content moderation, Safety classification

qwen-image-2512
MEDIUM

IMAGE GENERATION

qwen-image-2512

20-billion-parameter text-to-image foundation model. Generates highly realistic images, reduces the typical AI-generated look and improves text rendering inside images.

€ 0.0171 + VAT / image

Parameters: ~20B

Max context: N/A

License: Apache 2.0

Features: Image generation from text prompts

qwen-image-edit-2511
MEDIUM

IMAGE EDIT

qwen-image-edit-2511

20-billion-parameter image editing model. Performs creative edits while preserving subject identity and visual characteristics, with native output up to 2560x2560 pixels.

€ 0.0171 + VAT / image

Parameters: ~20B

Max context: N/A

License: Apache 2.0

Features: Image editing with prompts, Multi-image input

Meta

Qwen

OpenAI

DeepSeek

BAAI

Mistral AI

Microsoft

Google

LOG IN TO START Learn about our pricing

Examples of what you can do with our AI APIs

Enterprise RAG systems

Create AI assistants that answer using documents, knowledge bases and company content, such as technical manuals, product documentation or support content for operators.

Virtual assistants

Integrate conversational AI models to create chatbots, copilots and virtual assistants that support customer care by answering customers’ questions and guiding them through websites, portals and interfaces.

Semantic search

Improve semantic search across repositories, intranets and B2B portals with embedding endpoints and advanced search, even when queries use different words from those found in the documents.

Visual analysis

Extract usable text data from visual inputs such as images, PDFs, scanned documents, screenshots, charts, layouts and tables.

Text generation and summarisation

Use the endpoints to generate, rewrite, transform and summarise content such as product descriptions, emails, tickets, reports, meeting minutes and long documents.

Data classification and extraction

Automate the reading and classification of text and support tickets, as well as the extraction of useful information from reviews, surveys or editorial content.

Translation and localisation

Integrate machine translation capabilities into platforms, editorial workflows, portals and business tools to manage content, documents and support activities in multiple languages.

Images and creative content

Generate visuals, concepts and graphic assets from prompts for campaigns, catalogues, prototypes, websites and e-commerce.

Grow your AI projects within the Aruba Cloud ecosystem

This service gives you access to AI models and can integrate with a broader ecosystem of tools and services to maximise the use and benefits of AI for your business. Among them, AI Tools helps you add ready-to-use AI capabilities to your applications, platforms and processes.

AI Tools

Ready-to-use AI tools that support specific business activities, operations and day-to-day workflows, going beyond a simple response generated by an LLM.

Find out more

GPU on Demand

Boost GPUs in your Kubernetes environments for AI, machine learning, rendering, analytics workloads, and applications requiring accelerated computing capabilities, only when needed.

Find out more

Dedicated GPU servers

Physical servers with GPUs for AI workloads, inference, Machine Learning and projects that require dedicated resources, configurability, isolation and infrastructure control.

Find out more

European infrastructure ready to grow with your AI project

From API endpoints to dedicated architectures, Aruba Cloud provides a scalable ecosystem on proprietary infrastructure, with European data governance, composable services and specialist support.

Private data centers

Develop your AI projects on Aruba proprietary infrastructure, with geographic proximity, operational continuity and the capacity to grow at enterprise scale.

European data governance

Keep your data on European infrastructure, with greater control over management, protection and regulatory requirements, without it being used to train models.

From testing to production

Start with API integration, validate your use case and expand the project with GPUs, dedicated servers, private cloud and custom services based on your business needs.

Predictable costs

Choose when to use pay-as-you-go resources and when to move to dedicated infrastructure so you can scale AI with a model that is more sustainable over time.

Frequently asked questions about Aruba AI Models

  • AI Models is the Aruba Cloud service that provides access to AI models through API endpoints. You can integrate artificial intelligence capabilities into applications, workflows and digital services without directly managing the inference infrastructure.

  • To use the API endpoints, create an Aruba Cloud account or sign in with your existing account. Make sure you have available credit, then generate an API key from the Aruba Cloud Management Platform to make calls to the available endpoints.

  • No. API endpoints allow you to use AI models through APIs without configuring servers or GPU infrastructure. For projects that require greater control, Aruba Cloud also offers GPUs, dedicated servers and private cloud environments.

  • No. Data processed through AI Models is not used to train the models. This helps you integrate AI into business processes while maintaining greater control over the information being processed.

  • Yes. The endpoints are designed to integrate with applications and workflows through API calls compatible with OpenAI standards. This simplifies service adoption in existing environments and reduces integration work.

  • The service is designed for companies, IT teams, developers, software companies and system integrators that want to integrate AI capabilities into digital products, internal applications, business workflows or customer-facing services.

  • Yes. You can start with a limited use case, validate its value and then extend the integration according to the project’s needs. The service is suitable both for experimentation and for developing production applications.

  • API endpoints let you use AI models through APIs without managing the infrastructure. Dedicated GPU servers are suitable when you need dedicated physical resources, greater control over the environment or specific AI workloads on dedicated infrastructure.