Google Vertex Gemini 3 icon

Google Vertex Gemini 3

Send messages to Gemini 3 on Google Vertex AI

Google Vertex Gemini 3

AI-generated

Summary

Send a Gemini 3 request to Google Vertex AI using provided prompts and settings.

Inputs

  • projectId (required) — Google Cloud project ID to use.
  • modelName — Gemini model identifier; empty to auto-select the latest flash model.
  • thinkingLevel — Reasoning level (MINIMAL, LOW, MEDIUM, HIGH) or expression.
  • messages — Collection of role-text message pairs defining the conversation context.
  • systemInstruction — Optional system-level instruction for the model.
  • stream — Whether to use streaming endpoint; result is aggregated before passing downstream.
  • responseFormat — Output format, either 'text' or 'json'; determines if responseSchema applies.
  • responseSchema — JSON schema to force the model output to conform to when responseFormat is 'json'.
  • safetySettings — Per‑category content blocking thresholds.
  • options — Additional generation options such as temperature, topK, topP, maxOutputTokens, grounding.

Output shape

An array of output records; each record contains a json field with the shaped Gemini response (text, optional thoughtSummary, parsedJson, etc.).

The operation always returns a list; when streaming is enabled the list contains a single aggregated record. Errors may result in a record with an error field if continueOnFail is enabled.

Examples

Example 1: Simple text generation using the latest flash model

Project ID set, Model left empty, single user message "Hello", Response Format = text

Example 2: Structured JSON output with safety settings

Response Format = json, Response Schema provided, Safety Settings set to block low harassment, stream disabled

Example 3: Streaming response with grounding

Enable Google Search Grounding, Stream Response = true, Project ID and Model specified, single message prompt

Links

Discussion