ai api
Replicate
Replicate provides hosted APIs for running public machine-learning models and tools to deploy custom models. Public models use model-specific billing, commonly based on processing time and hardware, while some are priced by input or output units.
Data status: partially_verified · Checked: 2026-10-16
Good fit if
Run public models through an HTTP API and client libraries
May not fit if
There is no universal inference rate; cost depends on model, hardware, runtime, or input/output quantity.
Features and benefits
- Run public models through an HTTP API and client libraries
- Deploy custom models using Cog
- Model pages provide model-specific usage estimates
- Create predictions asynchronously and receive updates through webhooks
Limitations
- There is no universal inference rate; cost depends on model, hardware, runtime, or input/output quantity.
- Most private deployments bill for the full time dedicated hardware is online, including setup and idle time; fast-booting fine-tunes are an exception.
- Predictions created through the API have inputs, outputs, and logs removed after one hour by default.
Pricing
Public models
Price unknown · usage_based
Model-dependent. Many models are charged by hardware runtime; others use input/output units. Example: FLUX Schnell is listed at $3.00 per 1,000 output images. No general free quota is stated on the pricing page.
Pricing sourcePrivate deployments
Price unknown · usage_based
Hardware-specific per-second charges apply while most dedicated deployments are online, including idle time. Exact price depends on the selected hardware and deployment.
Pricing sourceSources
- https://replicate.com/ · 2026-10-16: hosted model platform
- https://replicate.com/pricing · 2026-10-16: billing models, example model price, hardware rates, private deployment billing
- https://replicate.com/docs/reference/http · 2026-10-16: HTTP API, predictions, webhooks
- https://replicate.com/docs/topics/billing · 2026-10-16: prediction data retention