A Gemini API key is a credential that authenticates your calls to Google's Gemini models: each request carries the key in the Authorization header, and usage is metered and billed by token. There are two ways to get a working key: Google AI Studio (official), which issues a key for Gemini only, or Kunavo, where a single OpenAI-compatible key reaches Gemini, Claude, GPT, and image, video, and audio models — without a Google Cloud project. This guide shows how to get a key through either method and make your first call.
Option 1 — Google AI Studio (official)
This is the right path if Gemini is the only model you plan to call.
- Sign in to
aistudio.google.comwith your Google account. - Open Get API key → Create API key. The key is in the format
AIza.... - The free tier is limited and suited to prototyping. For production volume, the key must be linked to a Google Cloud project with an active billing account.
If you also use Claude, image, or video models, this path means managing separate keys, SDKs, and bills for each provider.
Option 2 — Kunavo (recommended for production)
Kunavo puts Gemini behind an OpenAI-compatible API. One sk-kn- key calls Gemini and every other model, with a single wallet billed via Stripe and no Google Cloud project at all. Gemini costs around 70% below Google's official list prices — Flash, for example, costs $0.09 per 1 million input tokens and $0.75 per 1 million output tokens. It's pay-as-you-go starting with a minimum $10 top-up, and your balance never expires.
- Sign up at kunavo.com.
- Top up $10 — pay as you go; your balance never expires.
- Create a key at
/app/keys. The key starts withsk-kn-and is shown only once; save it in an environment variable. - Make a call by pointing the OpenAI SDK to
https://api.kunavo.com/v1and using a Gemini slug.
Quickstart
Keep using the official OpenAI SDK. Change base_url and the model name — nothing else:
import os
from openai import OpenAI
client = OpenAI(
api_key=os.environ["KUNAVO_API_KEY"],
base_url="https://api.kunavo.com/v1", # dòng duy nhất thay đổi
)
resp = client.chat.completions.create(
model="gemini-2-5-flash",
messages=[{"role": "user", "content": "Giải thích RAG trong hai câu."}],
)
print(resp.choices[0].message.content)Key security
- Never expose your key on the client side. Browser and mobile bundles are public — always proxy requests through your own server. Store the key in an environment variable, outside version-controlled code.
- Use one key per project or environment, so a single leak can be revoked without taking everything else down.
- Rotate keys regularly. Kunavo allows multiple active keys at once, so rotation is a smooth transition rather than a hard cutover.
- Set a spending limit for each key in the dashboard and monitor
/app/usageto flag unexpected traffic, broken down by key.
Frequently asked questions
Is the Gemini API free?
Google AI Studio issues a key on a free tier with low rate limits, enough for prototyping. Production traffic moves to paid billing on Google Cloud. Kunavo is pay-as-you-go starting with a minimum $10 top-up, requires no Google Cloud billing account, and the balance never expires.
Can I get a Gemini key without a Google Cloud account?
Yes. With Kunavo, you create a single sk-kn- key and call Gemini through an OpenAI-compatible endpoint, with no need to create a Google Cloud project or enable billing. Google AI Studio also issues a free-tier key without a project, but Google requires a billing account for production.
What's the difference between a Google Gemini key and a Kunavo key?
A Google key calls only Gemini and is billed through a Google Cloud account. A Kunavo key calls Gemini along with Claude, GPT, image, video, and audio models through a single OpenAI-compatible API, with one wallet billed via Stripe. For model-by-model pricing and cost examples, see the Gemini API pricing guide.
Which Gemini model should I start with?
Start with gemini-2-5-flash for everyday chat, extraction, and summarization — it's fast and extremely inexpensive. Switch to gemini-3-1-pro when you need harder reasoning, vision, and long-context analysis.