> For the complete documentation index, see [llms.txt](https://docs.parasail.io/parasail-docs/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.parasail.io/parasail-docs/products/capacity.md).

# Elastic Endpoints

Dedicated GPU performance with per-token billing. Learn about Elastic Endpoints pricing, sales-led setup for selected models, and provisioned throughput.

Elastic Endpoints provide reserved inference throughput for production workloads without requiring you to manage infrastructure or orchestration software.

Each Elastic Endpoint is private to your organization. You don't share the endpoint with other customers, and your traffic doesn't compete on a public model endpoint. Parasail manages the infrastructure behind it and provisions capacity for your workload based on the model, traffic shape, latency goals, and service agreement.

For eligible models, you pay for tokens without a GPU-hour commitment or an idle-GPU bill.

## Choose Elastic Endpoints

Use Elastic Endpoints when you want private, autoscaling capacity for an eligible model without managing GPUs or paying by GPU-hour.

* [Serverless](/parasail-docs/products/overview.md) provides shared, on-demand inference for hosted models.
* **Elastic Endpoints** provide private capacity with per-token billing for eligible models.
* [Dedicated Instances](/parasail-docs/products/overview-1.md) provide control over model, hardware, and scaling, with GPU-hour billing.

## Pricing and model availability

Elastic Endpoints are available for selected models, with availability expanding model by model. Availability on Serverless doesn't mean a model is eligible for Elastic Endpoints.

Elastic Endpoints have one starting pricing tier, **Serverless + 25%**, based on the corresponding Parasail Serverless model rates:

* **Input tokens:** 1.25 × the corresponding Serverless input-token rate.
* **Output tokens:** 1.25 × the corresponding Serverless output-token rate.

Use Parasail's shared Serverless rates as the baseline for this pricing. There is no GPU-hour commitment or idle-GPU bill for eligible models. See [Pricing](/parasail-docs/billing/pricing.md) for the Serverless rates and the other products' billing models.

## Get started

[Contact sales](https://www.parasail.io/contact) to confirm model eligibility and set up your Elastic Endpoint. Sales is required for initial setup.

After setup, use your endpoint's status page to review its capacity and rate limits.

Elastic Endpoints expose reserved serving capacity through a managed API endpoint: you get token-based billing and a provisioned throughput target, while Parasail handles placement, scaling, and reliability.

## How capacity works

Parasail defines Elastic Endpoint capacity during setup instead of using a public, fixed rate-limit tier. Parasail sizes the endpoint for the throughput your workload needs and adjusts the managed backend behind the endpoint as traffic changes.

The Elastic Endpoint status page focuses on operational capacity signals:

| Status field             | What it means                                                                                                   |
| ------------------------ | --------------------------------------------------------------------------------------------------------------- |
| **Current capacity**     | The serving capacity available to the endpoint right now. It changes as Parasail scales the managed backend.    |
| **Provisioned capacity** | The sustained throughput target configured for the endpoint. The endpoint's plan or agreement sets this target. |
| **Request activity**     | The live request volume using the endpoint, shown so you can compare current traffic with available capacity.   |

If traffic exceeds the current available capacity, the endpoint can return HTTP 429 errors while the backend catches up. If traffic exceeds the provisioned capacity for a sustained period, contact Parasail to increase the endpoint's throughput target.

## Autoscaling behavior

Parasail autoscaling smooths backend changes to protect quality of service and infrastructure stability. That smoothing means the endpoint can absorb normal variation, but sudden traffic spikes can temporarily exceed available capacity before scaling catches up.

For steady growth, Parasail can increase the provisioned capacity for the endpoint. For unexpected bursts, clients should retry 429 responses with backoff and idempotency controls.

The "Number Running Requests" tab on the Elastic Endpoint status page shows a graph of live request activity.

## Next steps

* [Dedicated Instances overview](/parasail-docs/products/overview-1.md)
* [Auto-Scaling](/parasail-docs/products/overview-1/auto-scaling.md)
* [Rate limits](/parasail-docs/operate-in-production/limits-and-quotas.md#rate-limits)
* [Pricing](/parasail-docs/billing/pricing.md)