Skip to main content
Register
On-Demand Inference

Private Preview: Serverless AI inference with pay-per-token pricing on a sovereign European cloud.

Access a curated catalog of AI models through an Inference API with usage-based, pay-per-token pricing on Exoscale's sovereign European cloud. No retention of prompts or model responses, and no training on your data.
Join the Waitlist
Private Preview

On-Demand Inference is available by invitation in AT-VIE-2. During the Private Preview, no SLA, availability commitment or service credits apply, and models, limits and prices may change. The current model catalog is not final and may expand during this phase.

Exoscale’s On-Demand Inference


On-Demand Inference is our usage-based, serverless AI inference offering. It gives you direct access to a curated model catalog through an OpenAI-compatible Inference API, enabling you to integrate AI features quickly.

There are no instances to size and no idle costs. Connect to the API and pay only for what you consume. This operating model is well suited to rapid prototyping, unpredictable traffic, and teams that want immediate AI capabilities on a secure, sovereign European cloud.

The service is currently available in Private Preview.

Why Choose On-Demand Inference on Exoscale?

Transparent European location

Transparent European location

Private Preview models currently run in AT-VIE-2 in Vienna, Austria. For General Availability, Exoscale plans service locations in the EU and Switzerland.

Curated and expanding model catalog

Curated and expanding model catalog

Use models for chat completions and embeddings today. The Private Preview catalog is not final and can expand with additional models and categories.

Zero Data Retention

Zero Data Retention

Prompt and model-response content is processed without being stored after inference. Exoscale retains only metadata required for usage reporting and does not use customer inputs or outputs to train, retrain, or fine-tune AI models.

Usage-Based Pricing

Usage-Based Pricing

Pay exclusively for what you use. No GPU sizing, no capacity planning, no infrastructure operations.

Clear model lifecycle

Clear model lifecycle

Get transparency on model availability and lifecycle so you can plan adoption, integration, and future updates with more confidence.

Integrated with the Exoscale platform

Integrated with the Exoscale platform

Connect AI as a Service with the rest of your sovereign Concrete AI portfolio, including Vector databases, Storage, and Kubernetes.

Data Handling and Service Commitments

On-Demand Inference keeps customer content transient. Prompts and model responses are not retained after inference, prompt text itself is never stored in the cache, and customer inputs and outputs are not used to train, retrain, or fine-tune AI models. Exoscale retains only the usage metadata required for billing and statistics.

During the Private Preview, processing takes place in AT-VIE-2 in Vienna and no SLA, availability commitment, or service credits apply. An SLA is planned for General Availability.

On-Demand Inference
Data retention OFF ALWAYS Prompts and responses are not stored after inference
Model training and fine-tuning OFF ALWAYS Customer data is not used
Processing in Europe ON Private Preview: AT-VIE-2, Vienna
Prompt text caching OFF ALWAYS Prompt text is never stored in the cache
SLA at General Availability ON Private Preview: no SLA

Private Preview Tentative Pricing

Get started with the models currently available in the Private Preview and pay only for the tokens you use, without sizing GPUs or reserving capacity. The catalog will continue to grow during this phase, giving you more options as new models become available.

See the technical documentation for current model details and service boundaries.


Model Input per 1M tokens (EUR / USD / CHF) Output per 1M tokens (EUR / USD / CHF) Cache read per 1M tokens (EUR / USD / CHF) Availability
Qwen3.8-27B 0.40 2.40 0.09 Preview
Qwen3-Embedding-8B 0.09 Not applicable Not applicable Preview
  1. [1]

    Prices are per 1 million tokens. Prices are tentative, and Exoscale reserves the right to adjust them during or after the Private Preview.

  2. [2]

    Cache-read pricing lowers the cost of eligible input tokens reused during conversations.

  3. [3]

    Private Preview is available by invitation in AT-VIE-2. No SLA, availability commitment, or service credits apply. See the service boundaries.

Designed for common AI workloads

Use On-Demand Inference to validate ideas quickly and serve applications whose traffic changes over time. Pay by token while demand is uncertain, and move predictable or mission-critical workloads to Dedicated Inference later.

Rapid prototyping & experimentation


Test models, prompts, and use cases quickly without provisioning infrastructure. Ideal for early-stage projects and proof-of-concept. Move from an idea to a working API integration while keeping experimentation costs aligned with actual usage.

Variable or unpredictable traffic


Handle bursty workloads and changing demand patterns with token-based consumption, without committing to fixed GPU capacity. This makes it suitable for pilots, seasonal demand, and applications whose usage has not yet stabilized.

Multi-team or decentralized usage


Provide a shared AI capability across teams via API access, without requiring each team to manage infrastructure or deployments. Scoped AI API keys help teams share the service while maintaining clear access boundaries.

Gradual path to production


Start with managed models and move to dedicated deployments when workloads become predictable or mission-critical. The OpenAI-compatible API helps keep your application integration consistent as the deployment model evolves.

On-Demand Inference Features

Join the On-Demand Inference Private Preview

Join the waitlist to evaluate the current model catalog and help shape the service before General Availability.

Join the Waitlist

Combine On-Demand Inference with…

Build a complete AI application stack by combining On-Demand Inference with Dedicated Inference for predictable workloads and managed vector databases for retrieval-augmented generation and semantic search.

Dedicated Inference

Dedicated Inference

Fully managed, secure, and production-ready API endpoints for any open-weight AI model.

Discover
Managed pgvector

Managed pgvector

PostgreSQL with pgvector extension. Perfect for hybrid workloads needing both relational and vector data.

Discover
Managed Vector search

Managed Vector search

OpenSearch-based vector database. Optimized for pure AI workloads and large-scale semantic search.

Discover

On-Demand Inference Documentation

Explore technical documentation and practical guides for connecting to On-Demand Inference, creating AI API keys, and controlling model access.

Quick Start

Create an AI API key and send your first request to an OpenAI-compatible endpoint.
Get Started

How-to Guides

Follow practical instructions for managing model access and using On-Demand Inference with supported tools.
Explore How-to Guides

Explore More Exoscale Services

Boost your On-Demand Inference solution by adding complementary offerings to help you achieve greater availability and performance, as well as expert support for any workload. Exoscale has the right service to support your project’s growth.

Kubernetes

Scalable Kubernetes Service

Deploy containerized applications on a production-ready Kubernetes cluster in under two minutes. Use SKS as the control layer for your virtual machine instances, with support for CLI, API, Terraform, and other DevOps tools.

Discover
object-storage

Simple Object Storage

Use a highly scalable and S3-compatible storage solution for unstructured data. Ideal for storing backups, logs, static assets, or media, fully integrated with Exoscale regions and access-controlled via API.

Discover
Support Plans

Support Plans

Get the help you need to run your infrastructure with confidence through flexible support plans, designed to provide expert guidance, faster response times, and dedicated assistance tailored to your business.

Discover

Frequently Asked Questions about On-Demand Inference

What is an On-Demand Inference service?

On-Demand Inference is Exoscale’s serverless Inference API, also commonly described as Model-as-a-Service (MaaS). It provides pay-per-token access to a curated catalog of AI models without requiring you to provision GPUs or operate model-serving infrastructure.

Does On-Demand Inference support Zero Data Retention (ZDR)?

Yes. Exoscale does not persist inference request payloads or model responses after processing, and prompt text itself is never stored in the cache. Only metadata required for usage reporting is retained, and customer prompts, inputs, and responses are not used to train, retrain, or fine-tune AI models. This guarantee covers the Exoscale inference service; customer applications and connected tools may have separate logging or retention behavior.

How is On-Demand Inference different from Dedicated Inference?

On-Demand Inference is designed for flexible, usage-based consumption (you pay per token) of a curated model catalog. Dedicated Inference is designed for customers who want to deploy and operate their own chosen model on managed, dedicated GPU-backed infrastructure.

Can I choose from multiple model categories?

Yes. On-Demand Inference currently provides models for chat completions and embeddings. The catalog can expand with additional models and categories during the Private Preview and beyond.

Can I use my existing AI tools with On-Demand Inference?

On-Demand Inference offers an OpenAI-compatible API. Tools that let you configure the API endpoint and credentials can connect without changing their core request format. See the Quick Start for currently supported requests and parameters.

How does the token pricing work?

Pricing is calculated based on the number and type of tokens processed. Input, output, and eligible cache-read tokens can have different rates. Cache-read pricing lowers the cost of eligible input tokens reused during a conversation, while prompt text itself is never stored in the cache. Cache-write support, including TTL configuration, is coming soon. See the technical documentation for current tentative pricing and cached-token details.

Can I monitor and control On-Demand Inference usage?

During the Private Preview, you can monitor On-Demand Inference usage in the Exoscale Portal. More granular controls, including limits per AI API key, are coming soon.

Where will the On-Demand Inference solution run?

During the Private Preview, On-Demand Inference is available in Vienna, Austria, in the AT-VIE-2 zone. For General Availability, Exoscale plans an SLA and availability in an EU location and a Swiss location. See the technical documentation for current location availability.