Replicate

★ 4/5 · Infrastructure/Database

Replicate is a hosted API to run open and commercial AI models (image, video, LLM, audio) with pay-per-second/token billing and an official model catalog.

Pros

Huge public model zoo. Official pricing is usage-based at replicate.com/pricing (hardware per-second + some models per-token). No monthly seat tax required to start.

Cons

Costs spike with GPU time
you must read each model's hardware. Not a training platform. Cold starts on some models.

Cost

Pay-as-you-go only (no monthly seat plan). Hardware on replicate.com/pricing: CPU $0.000100/sec; Nvidia T4 $0.000225/sec; L40S $0.000975/sec; A100 80GB $0.001400/sec; H100 $0.001525/sec. Some public models billed per output (example printed: flux-1.1-pro $0.04/image).

Verdict

Hosted inference catalog for prototypes and production sidecars.

Visit Replicate