Any model\, one API\, billed by the second.
Replicate runs thousands of open models behind a consistent API with no infrastructure to manage\, charging per second of compute — from $0.000225 on a T4 to $0.001525 on an H100.
Per-second billing with no idle cost is exactly right for bursty workloads: you pay for what you run and nothing when you don't. For steady high volume\, dedicated hardware is eventually cheaper. Cog lets you deploy your own models the same way.
Quick Information
Platform
Web
Pricing
Usage-based (from $0.000225/sec)
API
Available
Category
Coding & Development
Pros and Cons
Pros
- Thousands of models behind one API
- Per-second billing with no idle cost
- No infrastructure to manage
- Deploy custom models with Cog
- Transparent published rates
Cons
- No free tier
- Costs add up on sustained workloads
- Cold starts on less popular models
- Less control than self-hosting