Plain terms for API access to Inference.
We provide OpenAI-compatible inference endpoints for the models published at /v1/models. Capacity is best-effort on a single GPU node per region; we migrate capacity between regions to reduce cost.
No unlawful content, no attempts to bypass rate limits or authentication, no resale of keys without a written agreement. We may suspend keys that threaten service stability.
Usage is metered per token at the prices published on /pricing. Prices may change with notice; changes never apply retroactively. Monthly limits, where set, are enforced at request time.
The service is provided as-is. Single-GPU capacity can be preempted; we do not guarantee uninterrupted service and are not liable for indirect damages.
Our data practices are described in the privacy policy.