Novita AI

Novita AI is an inference cloud rather than a model of its own: one API in front of the order of 200 open models spanning text, image, audio and video, plus GPU instances and agent sandboxes for when you need the infrastructure underneath. The pitch is the unglamorous, useful part — no capacity planning, no cluster to babysit, billed by the token instead of the hour.

It publishes per-model token pricing rather than one blended rate, and the cheaper open models sit well under a dollar per million input tokens, which is what makes the trade-off legible against first-party APIs. It also treats machines as readers: the site directs agents to a documentation index and serves markdown when asked with an Accept header — a small decision, but a telling one for infrastructure aimed at developers.

Where it competes is breadth, price and convenience, not frontier quality. If you need the strongest model available it will be somewhere else, and hosting guarantees, regional data residency and support terms are the things to read before committing rather than the model count. For prototypes, batch jobs and cost-sensitive production workloads across open models, it is a reasonable baseline to price everything else against.