Masahiro Tanaka

Commented on Dorbit · Oct 16, 2026

Hi StackShortlist @Lars_759, Thank you for your interest! Large models take ~ 1 hour. Small ones (around 30B) are like 20min. We choose a node based on the network bandwidth and the model size. One good thing in Dorbit is that it doesn't charge you until the endpoint is ready.

Commented on Dorbit · Oct 8, 2026

Does anyone else find GPU pricing and availability for inference frustrating? I’m building Dorbit to make it easier to find a configuration that works for your model and your budget. Dorbit lets you run open models on dedicated inference endpoints, with a choice of GPUs and clear hourly pricing before you launch. That includes more affordable GPU options when you don’t need the fastest hardware. Different configurations offer different performance. The goal is to give you more choice in how much you spend and what you get. It’s still early, and I’d love your feedback. Which models are you trying to run, and what has been your biggest frustration with inference services? Happy to answer questions, and thanks for taking a look!