DeepInfra provides low-cost, OpenAI-compatible inference for open-weight LLMs and embedding models.
DeepInfra serves open-weight LLMs and embedding models through an OpenAI-compatible
API at a low per-token price, and can be used directly from Dify.
The Dify plugin runtime must be able to reach over HTTPS
(outbound port 443). The plugin calls DeepInfra's OpenAI-compatible endpoint at
; no other host is contacted. If your Dify
deployment uses an egress proxy, allow this host.
Select DeepInfra as the model provider in Dify, choose an available model, and use
it in applications, agents, or workflows.
To use a model released after this plugin version, choose Add Model, enter the
DeepInfra model ID exactly as it appears on the
DeepInfra models page, and set the context size and
capability flags to match that model.
Several DeepInfra models emit before their answer. With a small
budget the whole allowance can be spent reasoning, leaving the visible
response empty. Give reasoning models at least ~1000 max tokens.
This plugin sends the inputs required by the selected operation to DeepInfra. See
PRIVACY.md for details, and review
DeepInfra's privacy policy before use.
Last checked 10/23/2026 06:41 AM · version 0.0.1