What happened
Hugging Face released Inference Endpoints 2.0 on July 11 with auto-scaling, cold-start optimization and one-click deployment for more than 500 open-weight models. The update adds GPU sharing for bursty workloads and per-request billing down to the second.
Pricing starts at $0.60 per GPU hour for A10 instances. Hugging Face said the release targets teams that want self-serve open-model hosting without managing Kubernetes.
Why it matters
Self-hosting open models still requires ops expertise that many teams lack. Managed endpoints with auto-scaling lower the barrier while keeping model choice open.
Per-second billing also makes open models more viable for intermittent workloads.
MintedBrain take
Compare total cost against OpenRouter or Together AI for your traffic pattern before committing. Auto-scaling helps bursty loads but steady high volume may still be cheaper on dedicated infrastructure.
Discussion
Sign in to comment. Your account must be at least 1 day old.