Unexpected Training Credit Overuse and Inaccurate Usage Estimate

Describe your question/issue here! (delete this when you post)

  • Project Type: Instance Segmentation
  • Operating System & Browser: Windoiws 11 / Chrome
  • Project Universe Link or Workspace/Project ID: ball-segmentator
  • Do you grant Roboflow Support permission to access your Workspace for troubleshooting? (Yes/No): Yes

Hi everyone,

I ran into an issue with my project’s training usage that I hope someone from the Roboflow team can clarify.

My project doesn’t have Flex Billing enabled, but somehow the training usage exceeded the available and prepaid credits by a large margin. According to the platform’s own estimate before training SAM3, the expected usage was around 5 credits, yet the final consumption ended up being over 80 credits — and unfortunately, the resulting model wasn’t usable.

This big difference between the estimate and the actual usage caused me to overlook the training process, assuming it would stay within the expected limit.

I would really appreciate if the extra credits could be waived, and if the training usage estimation mechanism could be reviewed or improved to prevent similar situations in the future.

Thanks in advance for looking into this.