Shortly after launching its API in October, Runware announced $3 million in funding and documented its early inference stack. The platform served several image model families, including Stable Diffusion and FLUX, through a model catalog backed by Runware's own inference engine.
Runware
Also known as: Runware.ai · Runware Serverless · Sonic Inference Engine
Plain English
Runware offers hosted APIs for generating images, video, audio, and other content. Developers can call a model from its catalog or deploy their own code and containers on Runware Serverless. Runware builds modular data centers for part of the GPU capacity it serves.
Technical Overview
The Models API accepts tasks over HTTP or WebSocket, using a task ID to associate results with requests in a batch. Models have identifiers that include the publisher, model, and version. Runware Serverless runs custom workers on managed GPUs with scale-to-zero or reserved capacity, so teams can trade idle spend against startup delay and capacity guarantees.
Timeline (5)
2024
Milestone2025
MilestoneRunware announced a $13 million seed round led by Insight Partners. It described plans to extend its image-generation infrastructure into a broader media API, so applications could use one integration across image, video, and audio models.
2026
MilestoneRunware announced Sonic Pods, modular data centers designed for model serving. A pod packages GPU servers, local storage, networking, and closed-loop liquid cooling in a shipping container that joins Runware's inference network as one site.
- Release
Runware launched Serverless for customer models, code, and containers on the infrastructure behind its hosted API. Workers could use one, two, four, or eight GPUs and scale to zero, giving teams control over their serving code without operating their own GPU fleet.
- Release
Runware released a router that selects a model from the official catalog per request. A caller can set a price ceiling, describe required output features, and weight speed, quality, and cost without hardcoding a model ID.