Runware

Runware gives developers one API for more than 400,000 hosted, partner and community models, plus serverless and dedicated GPU capacity. It runs on modular data centers it builds itself and charges per output, which keeps generation prices low.

A single request can batch tasks across modalities, so an image, a video and an LLM chat can be sent together over REST or WebSockets, with a webhook per task if you want results pushed back. On open models you can control every parameter, including stacked LoRAs and a swapped VAE, and you can upload your own models, which stay private.

Pricing is listed per model, for example per million tokens, per image or per second of video. Failed or rate-limited requests are not charged, new business-email signups get $2 of credit, and high-volume users can ask for custom rates.