Self Hosted vs Roboflow Hosted

šŸ¤ Community Help
1

Hi,

We are running a fine tuned model that we trained on Roboflow. We have no issues open sourcing the model and data.

However, I want to know the difference between self hosting it vs Roboflow hosting it. I understand that the free tier has 10,000 calls per month. Does this include calls from both the ā€˜self hostedā€™ deployment as well as the roboflow hosted one? Is the use case for self hosted edge devices with no internet, reducing latency, data privacy etc?

If we self host the roboflow model on our cloud instance, what is the use case here? will we be incurring a cloud bill for no reason?

2

:information_source: Update (September 2026): Self-hosted inference no longer consumes credits. Running models on your own hardware with Roboflow Inference is free on every plan, including the Free Tier. Only Roboflow Cloud deployments (Serverless API, Dedicated Deployments, cloud video streams) use credits. See roboflow.com/pricing for current details.

Original answer (August 2024):

Hi @Abhishek - deployments are charged based on either the number of inferences run (if you are running your model on images) or the minutes of video stream processed (if you are running your model on video).

The limit and overage fees apply to both our Roboflow-hosted as well as self-hosted deployments.

Let me know if I can help with anything else!
Jacob