Call Volcengine Ark endpoints (Endpoint ID) from Dify. Supports LLM, text embedding, and speech-to-text.
Use Volcengine Ark endpoints ("Endpoint ID") in Dify.
This plugin is a good fit when you:
If you want to call Ark base models directly without creating endpoints, use the Volcengine Ark (Direct) provider instead.
Select (model version ) or
(model version ) for the matching endpoint,
and enter its Endpoint ID. Both models support text, image, and video
input, tool calls, and a 1M context window. DeepSeek supports up to 384K output
tokens; GLM supports up to 128K.
DeepSeek exposes the thinking switch and reasoning effort
/ / / (default ; disables thinking).
GLM always enables thinking and supports / / (default ).
Specifications checked on 2026-10-10:
DeepSeek V4.1 Flash,
GLM 5.3 Flash.
Token pricing uses DeepSeek's peak rate (RMB 2 input / 8 output per million tokens;
off-peak is RMB 1 / 4), and GLM's standard rate (RMB 0.8 / 2.8).
Select the base model matching your deployed endpoint:
Continue to enter your Endpoint ID. Seed 2.1 supports image/video input,
structured output, and reasoning effort from to (default ).
DeepSeek V4 GA supports text input and reasoning effort up to .
Seed 2.1 Lite also supports audio input via a public URL or Base64 data.
Its standard text/image/video token prices are RMB 0.8 input / 2.7 output per
million tokens. Audio input costs RMB 12 per million tokens; Dify's single
input-token price only estimates the standard input rate.
Seed 2.1 Lite specifications
checked on 2026-10-10.
Specifications and standard online token prices checked on 2026-09-07:
Seed 2.1 Pro,
Seed 2.1 Turbo,
DeepSeek V4 Pro GA,
DeepSeek V4 Flash GA.
Last checked 10/10/2026 10:50 AM · version 0.0.55