Overview

This plugin integrates the official and models
through the Chat Completions API. The display name follows the
stable API ID rather than a specific model version: DeepSeek may upgrade the model
behind that ID without requiring a new selection in Dify.

Both models support a 1,048,576-token context and up to 393,216 output tokens,
including reasoning. accepts text and images;
accepts text. Thinking is enabled by default, with effort; choose ,
, or , or disable Thinking Mode for a direct answer.
The entry defaults to a 65,536-token output limit.

For Flash, applies only in thinking mode, from 0.95 to 1.0;
applies only in non-thinking mode. For Pro, applies in non-thinking mode
and is ignored in thinking mode. See the
Chat Completions reference
and Thinking Mode guide.

The legacy IDs and remain
available with , so existing apps can keep their model selection.
They retain their 4,096-token output defaults and use the current Flash sampling
rules. The plugin sends the selected API ID unchanged. Select
when configuring the current Flash model explicitly.
Credential validation uses .

Images can be supplied as URLs or inline Base64 in user and tool messages.
Supported formats are JPEG, PNG, GIF, and WebP; see the
vision guide for request limits.
Local token estimates exclude images; the API's returned usage includes them.

Pricing varies by cache status and time window; refer to the
official pricing page.

Configure

Get an API key from DeepSeek, then
configure it under Settings → Model Provider.