Replicate
Replicate.com is a mature platform for deploying and sharing AI models.
CPU | 0 | 8 | 4 | 0.0001/s 0.36/h 0 | |
16 | 8 | 4 | 0.000225/s 0.81/h 0 | ||
48 | 16 | 4 | 0.000575/s 2.07/h -12% | ||
48 | 72 | 10 | 0.000725/s 2.61/h +12% | ||
40 | 72 | 10 | 0.00115/s 4.14/h -10% | ||
80 | 144 | 10 | 0.0014/s 5.04/h +10% | ||
384 | 680 | 48 | 0.0058/s 20.88/h |
There are large amounts of models made by the community as a showcase and for demonstration on the platform. You can publish your own model on Replicate for free, but you pay per second used by server.
With public models you only pay for the time the server is running, and with private models you also pay for the start up time.
You develop functions using the COG SDK, which makes it quite easy to containerize and deploy your AI models.
Replicate has stable support for webhooks, logging, and offer some observability.
The start up time for functions on Replicate is quite long (as of March 2024), so if fast responses are needed, it's better to use another vendor or configure Replicate to always keep a replica online.