Dify BetterToken setup: API key, Base URL, and custom models
Connect Dify to BetterToken with a Base URL, API key, and available model, then verify the connection and fix common errors. Practical guidance.
To connect Dify to BetterToken, prepare an API key, enter the Base URL shown below, and choose a current Model ID from the model plaza.
Key settings
| Field | Value |
|---|---|
| API Key | BetterToken API key |
| Base URL | https://bettertoken.ai/v1 |
| Model | A current Model ID from the model plaza |
Prerequisites
- The current version of the tool is installed
- A BetterToken API key: register and get one
- A Model ID from the model plaza
Additional requirements
- Dify Cloud or a self-hosted Dify instance with plugins enabled
- Permission to install model plugins and manage workspace models
Install
Install the OpenAI-compatible model plugin
Open Model Providers
In the Dify workspace, go to Integrations > Model Providers.
Install the official plugin
Click Install model providers, search for OpenAI-API-compatible, and install it.
Dify manages model providers as workspace-level plugins. After installation, apps with the required workspace permissions can select models from this provider.
Add a custom model
On the OpenAI-API-compatible card, click Add Model. If the card first shows Setup or Configure, open it and then choose the option to add a model.
Manual setup
Complete this setup in the Dify model-provider UI. You do not need to edit a configuration file.
| Field | Value |
|---|---|
| API Key | BetterToken API key |
| Base URL | https://bettertoken.ai/v1 |
| Model | A current Model ID from the model plaza |
Enter the model configuration
Use these values for the first model:
| Dify field | Recommended value |
|---|---|
| Model Type | LLM |
| Model Name | a model ID from the model plaza |
| Model display name | the same value as Model Name is fine |
| API Key | your BetterToken API key |
| API Base URL | https://bettertoken.ai/v1 |
| model name for API endpoint | the same model ID as Model Name |
| Completion mode | Chat |
| Model context size | match the selected model; keep the plugin default if unsure |
| Upper bound for max tokens | do not exceed the selected model; keep the plugin default for the first test |
| Compatibility mode | OpenAI compatible / strict |
| Token parameter name | Auto |
| Function Call Type | Not Support for the first test |
Use conservative values for the remaining optional capabilities:
- Thinking Mode Support: enable it only after confirming model support
- Stream function calling: select
Not Supportfor the first test - Vision Support: enable it only when both the model and endpoint accept image input
- Structured Output: keep it off until plain chat works
Keep Include Usage in Stream (stream_include_usage) at its default, Enabled. It adds stream_options: {include_usage: true} to request token counts in the final stream chunk; disable it only if the endpoint rejects that field. Stream mode auth (stream_mode_auth) defaults to Not Use; choosing Use makes credential validation use a streaming request, without changing normal generation settings. See the official provider fields.
Click Save. Dify validates the API Key, Base URL, and model ID.
Use the model in an app
Create or open a Dify app
Open a Chatbot, Agent, Chatflow, or Workflow.
Select the model
In the app model selector or an LLM node, choose the model you added under OpenAI-API-compatible.
Test plain chat first
Run the first request with a short text input and no tools or images:
Reply with exactly: Connection successfulEnable Agent capabilities afterward
After plain chat works, edit the model configuration if you need tool calls. Set Function Call Type to the mode supported by the model. OpenAI-compatible tool calling usually uses Tool Call, but you must verify support for the selected model and endpoint.
Verify the connection
Send a short test prompt. If the tool returns a response without an authentication or Model ID error, the connection works. Fully restart the tool after changing its configuration.
Switch models
Open the model selector or change the Model field in the configuration. Use the exact Model ID from the model plaza, then restart the current session.
Dify stores custom configuration per model ID. To add another model, click Add Model again under the same provider. Use the new ID for both Model Name and model name for API endpoint. You can reuse the Base URL and API Key.
Common errors
Saving returns 401
- Check that the API Key is complete
- Copy the exact Model ID from the model plaza.
- Check the key status and balance in the BetterToken Dashboard
Saving returns 404
- The API Base URL must be
https://bettertoken.ai/v1 - Do not append
/chat/completions - Make sure model name for API endpoint exactly matches the model plaza ID
Dify reports model not found
Model Name can affect how Dify displays and selects the model. The actual request model is controlled by model name for API endpoint. Use the same model ID in both fields.
Plain chat works, but Agent tools fail
Set Function Call Type back to Not Support and verify a basic LLM node. Change it to Tool Call only when the selected model supports tool calling, then test the tool arguments again.
Knowledge cannot use this model as an Embedding Model
This is expected. A chat LLM cannot replace an embedding model. This guide does not confirm BetterToken embedding or rerank endpoints for Dify; configure a supported embedding provider separately for Knowledge.
Advanced setup
Related guides
Technical details
Protocol, endpoint, and internal provider fields
This setup uses https://bettertoken.ai/v1. The tool appends the OpenAI-compatible endpoint path. Do not add /chat/completions or /responses unless a specific field explicitly requires it.