ModelRunner

ModelRunner is one API for huge curated list of AI models: images, video, audio, 3D and text.

One key, one request shape, every model. Trading a $0.35-a-second video model for one that costs $0.02 is a parameter change, not another integration.

Open weights and closed models, same catalog, same call. Google, OpenAI, ByteDance and Black Forest Labs sit right next to Stable Diffusion 3.5, Qwen-Image, FLUX.1 Kontext [dev], SDXL-Lightning, Florence-2 and LongCat-Video. If you've been renting a GPU to keep an open model running, you can put it head to head with a frontier model this afternoon and only write the call once. Nothing to fine-tune, nothing to deploy, and you keep the cheaper option when it wins.

Your results stay yours, and they're already an input. Everything you upload, send and generate lands in your asset library, hosted and indexed. Search it, filter by type or source, tag the keepers. Because the files are already hosted, any file field on any model can be filled straight from the library, so last week's output is this run's input with one click. Turn a phone photo into a clean packshot, the packshot into a video, the video into a 4K upscale, and never download or re-upload a thing. No bucket of your own to run.

It works from Claude, Cursor and ChatGPT. There's a hosted MCP server, so your assistant can search the catalog, read a model's input schema and run it inside the conversation. Sign-in is OAuth: a browser window opens once, and no API key ever goes into a config file.

Wrappers give you your own endpoint. Take a model, fix the prompt, expose only the two or three inputs your app actually needs, and call that. No backend to build or babysit. 15 are public already, doing things like on-model fashion shots, room redesigns and product packshots.

Every price is published before you run. Images from $0.012 a megapixel, video from $0.02 a second, quoted up front and locked to the request. No subscription, no idle GPU on your card, no minimum spend.

Sign up, top up, and run something in the playground. Then ship it with the REST API or the JavaScript and Python SDKs. MCP setup and the full tool list live at https://modelrunner.ai/mcp