Replicate — AI for Software Development

Replicate is a model-hosting API for developers who want to add image, video, speech, music or language models to an app without running GPUs. The homepage shows a few lines of Node, Python or HTTP code that call a model by name, and lists model categories such as image generation, speech, music, video and LLMs, including official models from several labs. Billing is pay-as-you-go. Most public models are billed by processing time on the hardware used, and some by input and output, for example $0.04 per output image for flux-1.1-pro or $3.75 per million input tokens for deepseek-r1. Setup and idle time on public models is free, but you may hit cold boots or scaling limits because you share a hardware pool. Private models run on dedicated hardware and you pay for setup, idle and active time, from $0.09/hr for a small CPU to $5.49/hr for an H100. Fast booting fine-tunes are billed only while active. You can push your own model with Cog, set up deployments and webhooks, and connect through an MCP server. The docs say select models run free, then you must set up billing, without stating a limit. Enterprise adds an account manager, higher GPU limits and volume discounts. If you need predictable monthly costs, per-second billing needs watching.