---
title: "Limits and Quotas - On-Demand Inference"
description: "Exoscale On-Demand Inference private-preview limits, available models, location, metrics, and tentative token pricing."
url: https://community.exoscale.com/product/concrete-ai/on-demand-inference/service-boundaries/limits-and-quotas/
product: "On-Demand Inference"
section: product
last_updated: 2026-10-22
---
> For AI agents: the documentation index is at https://community.exoscale.com/llms.txt. Every page is available as markdown at `<page URL>index.md` or with `Accept: text/markdown`.

# Limits and Quotas - On-Demand Inference

> [!IMPORTANT]
> **Private Preview**
>
> Functionality, model availability, limits, and prices can change during the
> private preview. The service is intended for evaluation and is not covered by
> an SLA or availability commitment. See [SLA - On-Demand Inference](https://community.exoscale.com/product/concrete-ai/on-demand-inference/service-boundaries/sla/index.md).

## Availability

On-Demand Inference is available in one zone during the private preview:

| Zone | Country | City | Availability |
| :--- | :--- | :--- | :---: |
| **`at-vie-2`** | Austria | Vienna | yes |

The OpenAI-compatible endpoint is
`https://inference.at-vie-2.exoscale-cloud.com/v1`.

### Inference API Endpoints

On-Demand Inference exposes OpenAI-compatible endpoints under the zone base URL:

`https://inference.<zone>.exoscale-cloud.com/v1`

| Capability | Method and path | Available model |
| :--- | :--- | :--- |
| Chat completions | `POST /v1/chat/completions` | `nvidia/Qwen3.8-27B-NVFP4` |
| Embeddings | `POST /v1/embeddings` | `Qwen/Qwen3-Embedding-8B` |

Authenticate inference requests with an AI API key in the HTTP `Authorization`
header:

```text
Authorization: Bearer <ai-api-key>
```

For request examples and supported model parameters, see the
[On-Demand Inference Quick Start](https://community.exoscale.com/product/concrete-ai/on-demand-inference/quick-start/index.md).

## Available Models

The initial private-preview catalog contains two public models. Both process
requests in AT-VIE-2 and are available under the
[Apache License 2.0](https://www.apache.org/licenses/LICENSE-2.0).

| Model | Status | Capabilities | Parameters and format | Context | License |
| :--- | :--- | :--- | :--- | :--- | :--- |
| [`nvidia/Qwen3.8-27B-NVFP4`](https://huggingface.co/nvidia/Qwen3.8-27B-NVFP4) | [Preview](https://community.exoscale.com/product/concrete-ai/on-demand-inference/overview/index.md#model-lifecycle) | Text generation; text, image, and video input; reasoning; coding; tool calling; structured output; streaming | 27B, NVFP4 | 262,144 tokens | [Apache-2.0](https://choosealicense.com/licenses/apache-2.0/) |
| [`Qwen/Qwen3-Embedding-8B`](https://huggingface.co/Qwen/Qwen3-Embedding-8B) | [Preview](https://community.exoscale.com/product/concrete-ai/on-demand-inference/overview/index.md#model-lifecycle) | Text embeddings; 100+ languages; instruction-aware retrieval; output dimensions from 32 to 4096 | 8B, BF16 | 32,768 tokens | [Apache-2.0](https://choosealicense.com/licenses/apache-2.0/) |

All private-preview models have the **Preview** lifecycle status. Model
capabilities, behavior, context limits, and availability can change during the
preview. Use `exo ai model show` to inspect the currently hosted version.

Custom model uploads and custom model weights are not supported by On-Demand
Inference. Use [Dedicated Inference](https://community.exoscale.com/product/concrete-ai/dedicated-inference/index.md) when you need to deploy a
different model.

## Access and Management Limitations

On-Demand Inference is managed through the Exoscale CLI and API during the
private preview.

An AI API key can grant access to all public models or be restricted to one or
more selected models. Availability is shared across customers and does not
reserve dedicated inference capacity.

## Usage and Billing Visibility

Usage and billing statistics provide visibility into request and token
consumption by model, AI API key, location, and billable token type. Billable
token types include input, output, and cache read where applicable.

## Tentative Pricing

Prices are tentative, in EUR/CHF/USD per one million tokens. Exoscale reserves the
right to adjust pricing during or after the private preview.

| Model | Input | Output | Cache read | Availability |
| :--- | ---: | ---: | ---: | :--- |
| `nvidia/Qwen3.8-27B-NVFP4` | 0.40 | 2.40 | 0.09 | Private preview |
| `Qwen/Qwen3-Embedding-8B` | 0.09 | Not applicable | Not applicable | Private preview |

Cache-read pricing lowers the cost of eligible input tokens reused from the
cache during a conversation. Cache-write support, including TTL configuration,
is coming soon.