IndexMapAnti-DoomscrollAbout
Company / HubImageVideoLLMCross-Modal

Replicate

Also known as: Replicate.com · Replicate Cog · Replicate Deployments

Plain English

Replicate is a hosted inference platform for generating images and video, processing audio, and running language models. Developers can call a catalog model through an API or package and deploy their own model with Cog, Replicate's open-source tool. The company is part of Cloudflare.

Technical Overview

Cog builds a container from a model's inference code and dependencies and generates an API schema from typed inputs. Replicate calls each invocation a prediction and supports asynchronous results through polling or webhooks. Deployments give a model a stable private endpoint with a chosen version, hardware, and scaling limits. Keeping instances warm reduces cold starts and incurs idle compute charges.

Timeline (7)

RSS
  1. 2022

    Milestone

    In its first blog post, Replicate described a hosted platform for sharing and running machine-learning models. It introduced Cog as a way to package inference code, weights, and dependencies so the same model could run on Replicate or on a team's own infrastructure.

  2. 2023

    Release

    Replicate introduced Deployments for public and private models. Each deployment has a fixed API endpoint, so an operator can change model versions, hardware, or the number of warm instances without changing the endpoint that the application calls.

  3. 2024

  4. 2025

    Release

    Replicate added official catalog model names to the same Predictions API endpoint that already accepted community model version IDs. A client could submit either kind of model to POST /v1/predictions through one request path.

  5. 2026

See also (8)

Open full graph
3 more connections