Gemini 3.5 Flash API Tutorial for Developers

Minimal request pattern

Use the official Gemini API docs as the source of truth for exact model IDs and SDK method names. The structure below is the important part for your app architecture:

import { GoogleGenAI } from "@google/genai";

const ai = new GoogleGenAI({
  apiKey: process.env.GEMINI_API_KEY,
});

const response = await ai.models.generateContent({
  model: "gemini-3.5-flash",
  contents: [
    {
      role: "user",
      parts: [
        {
          text: "Return JSON with features, risks, and nextActions from this release note.",
        },
      ],
    },
  ],
});

console.log(response.text);

If the current model ID differs in the docs, update only the model field and keep the same validation harness.

Production checklist

1. Store prompts in source control.
2. Validate structured output before parsing it into business logic.
3. Record failure examples in a small fixture set.
4. Add a fallback for rate limits, malformed output, and empty output.
5. Re-test prompts after model version changes.

Debugging notes

• If output is valid prose but invalid JSON, tighten the schema instruction and add server-side validation.
• If latency is high, reduce the input size and test whether the task needs an agent or only a direct model call.
• If quality varies, split the task into extraction, validation, and rewrite stages instead of one large prompt.