I’m testing different workflows that performs instance segmentation on an image using SAM3. I’ve noticed that when I run the exact same workflow on the exact same image, the cost is usually in the normal range, but sometimes—when the server seems slow, overloaded, or affected by some internal Roboflow issue—I get a timeout error.
What concerns me more is that in those cases I sometimes get charged around 10x the usual medium cost.
That seems strange and, honestly, not very fair. From my perspective, it looks like I’m being charged much more because of a slowdown or overload on Roboflow’s side. That would basically mean I pay more when the service performs worse, which does not really make sense.
I understand that workflow costs can vary somewhat depending on runtime, but a 10x increase is far beyond what I would normally expect.
Has anyone else experienced something similar? Am I misunderstanding why the credit usage suddenly jumps so much?
My understanding is that the credits charged for a run are based on how long the workflow runs. If that’s correct, what happens when Roboflow’s servers are overloaded? Is that extra time billed to the user, even when the slowdown is on Roboflow’s side?
- Project Type:instance segmentation (SAM3)
- Operating System & Browser:-
- Project Universe Link or Workspace/Project ID:
- Do you grant Roboflow Support permission to access your Workspace for troubleshooting? (Yes/No):yes