Gemini API quickstart: first call with curl and the Google GenAI SDK

One endpoint, the native Gemini protocol

To make your first Gemini API call through apiToken.sale, keep the Google protocol exactly as documented and change only two values: the base URL becomes https://router.apitoken.sale and the key is your apiToken.sale key, sent as the x-goog-api-key header. Every request and response keeps the native generateContent shape, so Google's documentation and SDK samples apply to the models and fields this endpoint supports. A few Google features are not available here: cachedContent and external file URIs are rejected (send files as inlineData), and the model list is limited to the published catalogue — the error table below lists each case.

One key and one prepaid balance cover every supported provider — Gemini alongside Claude and GPT. Gemini usage is metered at official Google token rates, and a flat 50% discount is applied before the cost is drawn from your balance. No Google Cloud project or billing account is involved on your side.

For Antigravity CLI, use the one-command setup: curl -fsSL https://apitoken.sale/setup/connect.sh | bash. On Windows PowerShell use irm https://apitoken.sale/setup/connect.ps1 | iex. It installs the native agy command and imports every Gemini model enabled for your key. No Google sign-in is needed.

See also: How to buy a Gemini API key

Create the key and list your catalog

1. 01Create a free apiToken.sale account and open the dashboard — no approval step or waitlist.
2. 02Generate one API key. It looks like sk-pool-… and works for Gemini, Claude and GPT alike.
3. 03Top up any whole-dollar amount by card or crypto; the prepaid balance does not expire.
4. 04Export the key as APITOKEN_API_KEY, then list the models your key can actually call:

curl https://router.apitoken.sale/v1beta/models \
  -H "x-goog-api-key: $APITOKEN_API_KEY"

Pick an explicit model ID from that response. gemini-3.6-flash is the right default for a first text call; a client library's built-in default may not be in the gateway catalog, and the router only serves the IDs it lists.

First call: generateContent with curl

curl https://router.apitoken.sale/v1beta/models/gemini-3.6-flash:generateContent \
  -H "x-goog-api-key: $APITOKEN_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"contents":[{"parts":[{"text":"Reply with exactly: connected"}]}]}'

The response is the standard Google shape: read candidates[0].content.parts and join the text parts. The same JSON carries usageMetadata with prompt, candidate and total token counts, so token- and cost-tracking code works from the very first call.

Before sending a large prompt, call :countTokens on the same model path. It returns a token count without generating anything — a free input estimate before you spend on generation.

Stream tokens with streamGenerateContent

curl "https://router.apitoken.sale/v1beta/models/gemini-3.6-flash:streamGenerateContent?alt=sse" \
  -H "x-goog-api-key: $APITOKEN_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"contents":[{"parts":[{"text":"Count from one to five"}]}]}'

The ?alt=sse query switches the response to server-sent events: each event is one incremental chunk in the same candidate structure, and the terminal event carries the aggregate usageMetadata. In the SDKs the same route is reached through generate_content_stream in Python and generateContentStream in JavaScript.

Stream anything user-facing so the first tokens render immediately. For batch jobs where only the final text matters, plain generateContent is simpler to parse and retry.

Official SDKs: Python and JavaScript

import os
from google import genai
from google.genai import types

client = genai.Client(
    api_key=os.environ["APITOKEN_API_KEY"],
    http_options=types.HttpOptions(base_url="https://router.apitoken.sale"),
)

response = client.models.generate_content(
    model="gemini-3.6-flash",
    contents="Reply with exactly: connected",
)
print(response.text)

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

const ai = new GoogleGenAI({
  apiKey: process.env.APITOKEN_API_KEY,
  httpOptions: { baseUrl: "https://router.apitoken.sale" },
});

const response = await ai.models.generateContent({
  model: "gemini-3.6-flash",
  contents: "Reply with exactly: connected",
});
console.log(response.text);

• Pass the bare base URL https://router.apitoken.sale; do not append /v1beta in SDK configuration.
• Pass a concrete model ID such as gemini-3.6-flash — never rely on a client default.
• Keep APITOKEN_API_KEY in the environment rather than in source code.

If every SDK request returns 404, check the request path for a doubled /v1beta/v1beta segment. The SDK appends its API version itself; configuring the host with /v1beta already included produces the doubled path.

What the first calls cost

Gemini calls settle at the exact official Google rate legs — input, cached input and output — and the flat 50% discount is applied on top. Post-discount prices per 1M tokens for the common text models:

Full Gemini rate card, including long-context rates →

Every supported model ID and price →

Troubleshooting the first response

Do not send Authorization: Bearer or the Anthropic x-api-key header on the native Gemini routes — x-goog-api-key is the only credential they accept. Because the wire format is unchanged, reverting to Google's own endpoint later is a one-line base-URL change.

Choose between Pro, Flash and Flash-Lite →

Frequently asked questions

Related guides