Replicate hosts a large catalog of open-source machine-learning models covering image generation, language, audio, video, and embeddings. Developers call models through a single HTTP API and pay by the second of inference compute, without managing their own GPU infrastructure.

The platform supports fine-tuning on custom datasets, private model deployments, and auto-scaling endpoints so that production traffic is served without manual capacity planning. Model publishers can also distribute their weights on Replicate and earn usage-based revenue.