Replicate
🛠️ Developer Toolspay-per-use
★4.3
Run open-source ML models via API
apimodelscloud
Try Replicate →Use Cases
- •Run open-source image generation models like Stable Diffusion and Flux via a simple API
- •Fine-tune language models like Llama on custom data without managing GPU infrastructure
- •Deploy community-contributed ML models as scalable API endpoints with pay-per-second billing
Integrations
Python SDKNode.js SDKREST APIGitHub ActionsZapierVercelNext.jsSwift SDK
Pros
- +Zero infrastructure management — run any model with a single API call
- +Pay-per-second billing means you only pay when code is actually running
- +Massive community model library with new open-source models added daily
Cons
- -Cold start latency can be several seconds when a model is not actively running
- -Costs can escalate quickly for high-volume or long-running inference workloads
- -Limited control over hardware selection and scaling compared to dedicated cloud GPU providers
Quick Start
1. Go to replicate.com and sign up with your GitHub account
2. Browse the Explore page to find a model (e.g., Stable Diffusion, Llama)
3. Try any model directly in the browser with the built-in playground
4. For programmatic access, get your API token from account settings
5. Run predictions via the REST API or Python client: replicate.run('model/name', input={...})
Pricing
Pay-per-use: Billed per second of compute time. CPU: ~$0.000100/sec. Nvidia T4 GPU: ~$0.000225/sec. Nvidia A40 GPU: ~$0.000575/sec. Nvidia A100 80GB: ~$0.001150/sec. Nvidia H100: ~$0.003200/sec. Some models billed per input/output token instead. No monthly minimum.