OpenRouter offers a free 1M-context multimodal model for agent work

OpenRouter has added Crouton Alpha, a free stealth model aimed at high-volume coding, tool use, document analysis and repeated agent tasks.

The model exposes a 1,048,576-token context window and accepts text, images, audio, video and files, with text output. OpenRouter lists input and output pricing at zero during the preview. OpenCode says the free window lasts about a week and is available only in selected regions.

OpenRouter describes Crouton Alpha as a lightweight reasoning model designed to scale out work that agents repeat all day. It supports tool calls and common reasoning controls through the OpenRouter API. The provider also says user data is not used for training.

The company behind the model has not been disclosed. Online guesses connecting it to Meta or other labs remain speculation, and there are no independent evaluations establishing quality or reliability. The model is also unavailable through OpenRouter’s EU in-region routing.

The practical appeal is clear: a million-token, fully multimodal endpoint at no cost can be used to test sub-agent architectures and document-heavy workflows without committing budget to a known flagship model. The tradeoff is uncertainty about the developer, long-term pricing and behavior after the stealth preview ends.