ModelRunner
Run any AI model without building a backend.
ModelRunner is one API for every AI model: images, video, audio, 3D and text. The interface never changes. Every model takes the same request shape, so which model you run is just a parameter. Something better ships next month and you point at it. A model gets too expensive and you drop to a cheaper one on the same call. Your code doesn't move, and you're never stuck with whatever you picked on day one.
What has shipped since
Nothing posted yet.1 comment
Newest first
Same prompt, two models, two bills. Qwen-Image ships its weights openly. You can download them and run them yourself, if you want to rent a GPU and keep it warm. Or you can call it here at $0.02 per megapixel, which works out to about two cents for a 1024 by 1024 image. GPT Image 2 doesn't ship weights. It's $0.145 per image, flat, whatever size you ask for. That's roughly 7x on the same prompt, and the interesting part is how little it costs you to find out which one you actually want. It's the same call either way. You change the endpoint string, keep everything else, and both results land in the same asset library. Open or closed is a parameter here, not an architecture decision.
- Launch week
- 31 Aug – 6 Sep 2026
- Placed
- #4 of 10
- Category
- Developer Tools
