Replicate

Cloud platform to run, fine-tune, and deploy open-source and custom machine learning models through a simple API, without managing GPU infrastructure.

Category

Machine Learning Infrastructure

Pricing

Usage-based, Enterprise

Target Market

Enterprise, Startups, Developers

Founded

2019

Headquarters

San Francisco, California, United States

• ✓
  Run models via API

  Access thousands of community-contributed, production-ready models (image, video, speech, music, and LLMs) and call them with as little as one line of code.

• ✓
  Cog

  Open-source tool for packaging machine learning models into containers that automatically generates an API server and handles cloud infrastructure.

• ✓
  Fine-tuning

  Customize existing models on your own data to create specialized versions for particular tasks.

• ✓
  Deployments & auto-scaling

  Deploy your own models on dedicated hardware with automatic scaling up and down based on traffic, paying only for active compute.

• ✓
  Client SDKs

  Official Python and JavaScript/Node.js client libraries for interacting with the platform programmatically.

• ✓
  Per-second usage billing

  Transparent per-second GPU and CPU pricing (plus per-token/per-image rates for certain models) so you are billed only for what you use.

• •
  Add AI features to applications

  Integrate image generation, video, speech, music, and language models into products through a single API without building ML infrastructure.

• •
  Deploy custom and fine-tuned models

  Package models with Cog and deploy them to production with managed, auto-scaling GPU infrastructure.

• •
  Prototype and experiment with ML

  Quickly test and iterate on open-source models in the cloud without provisioning or managing GPUs.

Pricing is usage-based. Most public models are billed by processing time at per-second hardware rates that vary by GPU/CPU tier (e.g. CPU ~$0.36/hr, Nvidia T4 ~$0.81/hr, L40S ~$3.51/hr, A100 80GB ~$5.04/hr, H100 ~$5.49/hr), with multi-GPU options scaling proportionally. Certain models (notably large language models and some image models) are billed per input/output token or per output image instead of by time. Private/custom models run on dedicated hardware and are billed for setup, idle, and active processing time. Enterprise adds a dedicated account manager, priority support, higher GPU limits, performance SLAs, and volume discounts. Exact per-model rates are listed on each model page and on the pricing page.