Gemini API

Paths

Gemini exposes two generation operations: GenerateContent returns a complete result, while StreamGenerateContent returns native Gemini streaming events. Both share Gemini authentication, model products, and upstream capability selection, but use different paths and delivery modes.

Generation paths must include the /v1beta/models/{model}: prefix. {model} may be gemini-2.5-flash or models/gemini-2.5-flash; the gateway normalizes the models/ prefix inside the model value. Model-list responses use the Gemini-style models/{publicModel} form.

Authentication

See Base URL and authentication and Headers for the complete header rules. Examples on this page use x-goog-api-key. Bearer is supported only for compatibility; x-api-key and the URL ?key= form are rejected.

Non-streaming request

curl https://dingjiai.com/v1beta/models/gemini-2.5-flash:generateContent \
  -H "x-goog-api-key: YOUR_USER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "contents": [
      { "role": "user", "parts": [{ "text": "Introduce the DINGJIAI gateway in one sentence" }] }
    ]
  }'

The request-body model and stream fields do not select the DINGJIAI model. The path {model} is authoritative.

Streaming request

curl -N https://dingjiai.com/v1beta/models/gemini-2.5-flash:streamGenerateContent \
  -H "x-goog-api-key: YOUR_USER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "contents": [
      { "role": "user", "parts": [{ "text": "List three integration notes" }] }
    ]
  }'

Streaming must use streamGenerateContent. Setting "stream": true in a generateContent request returns stream_not_supported.

usageMetadata and settlement

Both response modes prefer upstream usageMetadata; streaming keeps the last valid value. If usage is missing or invalid, generated text, thoughts, and functionCall output may contribute to estimation. A valid non-streaming token-priced success JSON object receives at least one output token when usage is missing or invalid, even when no output can be extracted (including {}). Streaming estimation requires extractable model output. Disconnects, cancellations, and delivery failures are not necessarily zero-cost. See Usage and billing for the complete rules.

Gemini image products do not use token usageMetadata; they are billed by the number of valid images actually returned. No valid image means failure settlement with zero charge.

Model list

curl https://dingjiai.com/v1beta/models \
  -H "x-goog-api-key: YOUR_USER_API_KEY"

The list contains Gemini products that are visible for the current entry, enabled, and have an available upstream candidate. Token products include generateContent and streamGenerateContent in supportedGenerationMethods; per_image products support only generateContent. Streaming support must therefore be read from each returned model product.