> For the complete documentation index, see [llms.txt](https://docs.roboflow.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.roboflow.com/deployment/roboflow-cloud/cloud-streaming/pricing-and-limits.md).

# Pricing and Limits

Understand Cloud Streaming credit usage, GPU sizing, price estimates, and processing limits before starting a run.

Cloud Streaming bills for each hour a Workflow runs on a GPU, at the rate of its GPU size. Usage is metered by the second. Every run counts: previews, tests, deployments, and uploaded videos. The bill does not change with frame rate, the number of frames processed, or how many people watch the output.

The table below lists the standard rate for each GPU size, also shown on the [credits page](https://roboflow.com/credits). Your Workspace's rates can differ (ex: on an Enterprise contract), and the app shows your rate before you start. To compare Cloud Streaming with per-image and Batch Processing billing, see [how each option bills](/deployment/choosing-a-deployment.md#billing).

## GPU Sizes

<table data-search="false"><thead><tr><th>GPU size</th><th>Credits per hour</th><th>API profile</th><th>Example models</th></tr></thead><tbody><tr><td>Small</td><td>0.1</td><td><code>gpu-s-v1</code></td><td>Object detection (ex: RF-DETR, YOLO26)</td></tr><tr><td>Medium</td><td>0.2</td><td><code>gpu-m-v1</code></td><td>SAM 2, or Workflows that run out of memory on Small</td></tr><tr><td>Large</td><td>0.4</td><td><code>gpu-l-v1</code></td><td>SAM 3</td></tr></tbody></table>

A deployment that runs all month (720 hours) uses 72 credits on Small, 144 on Medium, or 288 on Large at the standard rate.

The app and the Roboflow Agent pick the size from the Workflow's models, so you don't choose it. The [HTTP API](#http-api) can request a size.

## Web App

### Review the estimate

The source page shows an estimated hourly rate before you start. The "Cloud Streaming" option under "Deploy" in the Workflow editor (or "Use" in the Agent) shows the same estimate for a saved Workflow. The estimate does not reserve a GPU or lock the price.

"Start Processing" runs the Workflow as saved at that moment. If you edit it after you see the estimate, the GPU size and rate can differ, and the app asks you to confirm the new rate. Later edits do not change a running job.

While a job runs, its source page shows its current hourly rate, including jobs the Agent started. This is a rate, not total spend. To see spend, use [Credit Usage](#credit-usage).

Agent conversations, editor previews, stream lists, and shared output views do not show the rate, but their processing still bills. Promoting a preview shows its rate before you confirm.

### Automatic sizing

The app and the Agent pick the starting GPU size from the Workflow's models (ex: a Workflow with SAM 3 starts on Large). The size is not a frame-rate guarantee. If a run runs out of GPU memory, Roboflow stops it and restarts the same Workflow on the next larger size, when capacity and credits allow. The app shows the new rate when the replacement starts. Jobs never move to a smaller size. If a Workflow does not fit on Large, use a smaller model.

{% hint style="info" %}
The new-rate notice does not pause billing: the larger GPU is already running. "Confirm" closes the notice and "Stop" ends processing. Ignoring the notice or leaving the page does not stop the run.
{% endhint %}

Promoting a preview to a deployment keeps its GPU size and rate. See [Run a Workflow](/deployment/roboflow-cloud/cloud-streaming/run-a-workflow.md) for preview, deployment, restart, and promotion behavior.

## Credit usage

Models that run inside the job are covered by the hourly rate: they are not also billed per image. Blocks that call other paid services (ex: GPT, Claude, or Gemini blocks, or SMS and email notifications) bill at their own rates on the [credits page](https://roboflow.com/credits).

New starts, restarts, and automatic size increases need a Workspace that can spend credits. Running jobs stop when the Workspace reaches its credit limit. A job stopped this way stays stopped after you add credits: resolve billing, then start or restart it.

Open Workspace Settings, then "Credit Usage", to review actual consumption, date ranges, usage types, and API-key attribution where available. See [View Your Credit Usage](https://docs.roboflow.com/platform/billing-and-plans/view-credit-usage). The source page's hourly rate and Activity panel are not billing ledgers.

## Availability and access

Cloud Streaming is available to all Workspaces. You need access to the source and Workflow, and credits for processing. In the EU region, RTMP sources and webcam or phone sessions are not available yet.

"View Stream Processing" grants read access. Starting, stopping, watching, and managing sources need "Manage Source Video" or the corresponding API authority. Saved captures also require upload permission. See [permissions](/deployment/roboflow-cloud/cloud-streaming/manage-and-monitor.md#permissions).

## Processing limits

* A deployment has no time limit and bills until you stop it.
* A preview stops 2 minutes after its last viewer leaves (10 minutes if no one watches it) and after 1 hour at most. Webcam and phone previews cannot become deployments. See [Run a Workflow](/deployment/roboflow-cloud/cloud-streaming/run-a-workflow.md).
* Start time and frame rate are not guaranteed. They depend on available capacity and Workspace policy.
* Workflows must fit the largest GPU size and use blocks that run in the cloud. Test your models and blocks with a preview before you deploy.
* Camera FPS, processed FPS, and playback FPS measure different stages. A requested FPS cap is a limit, not a promise that every frame is processed or shown.
* Video playback depends on the camera codec, browser, and network. JSON-only Workflows do not need an image output. See [Consume Results](/deployment/roboflow-cloud/cloud-streaming/consume-results.md).
* Live results are temporary. To keep results, use a Workflow sink.

## HTTP API

A job you start with the API uses the GPU size you send, or Small (`gpu-s-v1`) if you send none. Unlike app starts, it is not sized from the Workflow and does not move to a larger GPU after a GPU memory error. Do not send app-only sizing or paid-validation fields.

Request field limits, authentication, and response shapes are defined in the [job operation](/deployment/roboflow-cloud/cloud-streaming/run-a-workflow.md#http-api). A request limit such as `maxFps` is not a throughput guarantee. See [Troubleshooting](/deployment/roboflow-cloud/cloud-streaming/troubleshooting.md) for credit, capacity, and resource failures.
