Getting the Free Gemini API in 2026 (3-min setup, no credit card)

AI 工具白嫖系列 — 把"免费但限速"的大模型 API 一次说清,顺便贴一张图给你看效果。

Google's Gemini API has had a free tier for a while, and in 2026 it's still the most generous
"no card required" option for indie devs. Here's the exact steps I use whenever I want a quick LLM
endpoint for a side project, scraping prototype, or Telegram bot.

1. Get the key

• Go to https://aistudio.google.com/app/apikey while signed in to any Google account.
• Click Create API key → Create key in new project.
• Copy the key. It starts with AIzaSy….

Heads-up: this key is bound to the project, not the account. Don't paste it into client-side
code — proxy it.

2. Test it (curl)

curl -s "https://generativelanguage.googleapis.com/v1beta/models/gemini-2.5-flash:generateContent?key=$KEY" \
  -H "Content-Type: application/json" \
  -d '{"contents":[{"parts":[{"text":"Say hi in 6 words"}]}]}' | jq .candidates[0].content.parts[0].text

You should get back something like "Greetings, fellow human explorer."

3. Free-tier limits that actually matter

Model	RPM	RPD	Notes
gemini-2.5-flash	60	1k	Best price/perf for chat.
gemini-2.5-pro	5	25	Slow but strong; great for eval.
gemini-2.0-flash	60	1k	Older; still fine for summaries.

Numbers are correct as of writing; Google's docs are the source of truth if it changes.

4. When the free tier is not enough

• Switch to paid tier (still cheap, $0.075/M input for 2.5 Flash) at the same page.
• Or pair it with a local Ollama + Llama 3.2 3B for fallback when you hit 429.
• Don't hammer with retries — the rate limiter will block you for ~10 min.

That's it. Three minutes, one key, one curl. If you want a ready-made Node/Python wrapper
to cache responses to SQLite, let me know in the comments — happy to share mine.

---

Tags: #ai #gemini #apitutorial #freestuff