Gemini Stream Generate Content

Receive streaming generation results over SSE with the Gemini Stream Generate Content endpoint.

Stream Generate Content uses the same Content, Parts, Bearer authentication, and generation configuration as standard generation, then continuously returns GenerateContentResponse chunks over one-way SSE. It is different from the bidirectional WebSocket-based Live API.

Clients should parse each SSE data: field at blank-line boundaries, then assemble text or Structured Output fragments in order. Do not call JSON.parse once on the entire response stream. Intermediate chunks might not contain finishReason; keep the latest usageMetadata when it appears. The same responseId associates every chunk in one generation. If candidates is missing, inspect promptFeedback.

POST/v1beta/models/{model}:streamGenerateContent

Use SSE to continuously return the GenerateContentResponse data block.

Authorization

bearerAuth

AuthorizationBearer <token>

Use Bearer your-api-key in the Authorization Header.

In: header

Path Parameters

model*string

Model ID used to generate content. The path automatically forms models/{model}.

Default"MODEL_ID"

Query Parameters

alt*"sse"

Select the SSE transfer format. This parameter is a URL query parameter and not part of the GenerateContentRequest body.

Default"sse"

Value in

"sse"

Request Body

Gemini uses Content and Parts to represent text, multimodal input, tools, and generation settings.

application/json

Text chat

Use user and model Content objects to organize single-turn or multi-turn text conversations.

contents[]: Conversation content in chronological order.
parts[].text: Text content in each message.
role: Message author; commonly user or model.

application/json

{
  "contents": [
    {
      "role": "user",
      "parts": [
        {
          "text": "Hello! Please introduce yourself."
        }
      ]
    }
  ]
}

Use Content objects and text Parts to organize single-turn or multi-turn text conversations.

contents*array<>

Text Content objects in chronological order. Each message provides content through parts[].text.

Items1 <= items

tools?array<>

tools that the model can use, such as functions, code execution, search, URL Context, File Search, MCP, and more.

toolConfig?

Configuration of tool call in requests.

safetySettings?array<>

Configure content masking thresholds by security category.

systemInstruction?

System commands set by developers, currently only supporting text content.

generationConfig?

Model generation and output configuration.

cachedContent?string

Name of the cached content used for context, formatted as cachedContents/{cachedContent}.

serviceTier?ServiceTier

The service level used by the request.

Value in

"unspecified"
"standard"
"flex"
"priority"

store?boolean

Control the logging behavior of the current request; After setting, it takes precedence over project-level log configuration.

Response Body

`text/event-stream`

`application/json`

candidates?array<>

Candidate answers generated by the model.

promptFeedback?

usageMetadata?

modelVersion?string

Model version used to generate responses.

responseId?string

Identify the ID of the current response.

modelStatus?

Lifecycle state of the underlying model currently used in response.

curl -X POST "https://example.com/v1beta/models/MODEL_ID:streamGenerateContent?alt=sse" \  -H "Content-Type: application/json" \  -d '{    "contents": [      {        "role": "user",        "parts": [          {            "text": "Please introduce quantum computing step by step."          }        ]      }    ]  }' \  --no-buffer

Text fragments from each event are concatenated in order; The same responseId indicates they belong to the same generation.

data: {"candidates":[{"content":{"parts":[{"text":"Quantum computing"}],"role":"model"},"index":0}],"modelVersion":"MODEL_ID","responseId":"response-stream-example"}

data: {"candidates":[{"content":{"parts":[{"text":"uses qubits to process information."}],"role":"model"},"finishReason":"STOP","index":0}],"usageMetadata":{"promptTokenCount":8,"candidatesTokenCount":18,"totalTokenCount":26},"modelVersion":"MODEL_ID","responseId":"response-stream-example"}

Google Gemini Stream Generate Content documentationReview the upstream streaming protocol definition and the latest field changes.

Gemini Stream Generate Content

Authorization

Path Parameters

Query Parameters

Request Body

Response Body

200text/event-stream

400application/json

401application/json

403application/json

404application/json

413application/json

429application/json

500application/json

503application/json

504application/json

`text/event-stream`

`application/json`

`application/json`

`application/json`

`application/json`

`application/json`

`application/json`

`application/json`

`application/json`

`application/json`