Is the Gemini API free to use? Yes, within limits

Google Gemini is one of the fastest and most capable model families around, but getting API access usually means creating a project, generating keys, and watching a billing dashboard. This guide shows you how to use the free Gemini API through DreamPrompting, a free, OpenAI-compatible gateway with automatic failover. Even better, it exposes something most wrappers hide: direct control over Gemini "thinking", so you decide when the model spends extra tokens reasoning and when it just answers.

Why a free Gemini API gateway

DreamPrompting sits in front of many providers and speaks the OpenAI Chat Completions format. That means any tool or SDK that already talks to OpenAI can call Gemini with a one-line base URL change. You do need an API key, and it is free: sign in with Google and copy one in a few seconds, no credit card. The limits are generous at 100 requests per minute per IP plus a rolling 24 hour account quota of 5,000 requests and 500,000 tokens, and if one upstream is busy the gateway fails over automatically. Under the hood this runs on Google's free AI Studio tier through the gateway, so please treat it as fair use rather than a production-scale pipeline.

• Base URL: https://dreamprompting.com/api/v1
• Auth: Authorization: Bearer YOUR_API_KEY (free from your account page)
• Chat: POST /api/v1/chat/completions
• Discovery: GET /api/v1/models and GET /api/v1/quota
• Models: google/gemini-3.1-flash-lite (default lite) or google/gemini-2.5-flash

The default call: thinking off, fast and cheap

On DreamPrompting, Gemini defaults to the lightweight gemini-3.1-flash-lite model with thinking turned OFF. This matters more than it sounds. When thinking is off you get fast, predictable, low-token answers, which is exactly what you want for classification, extraction, short replies, and high-volume calls where latency and cost add up. Here is the simplest possible request with curl. Swap YOUR_API_KEY for the free key from your account page.

curl https://dreamprompting.com/api/v1/chat/completions \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "google/gemini-3.1-flash-lite",
    "messages": [
      {"role": "user", "content": "Summarize what an API gateway does in one sentence."}
    ]
  }'

One free key, no thinking, no surprises. You can also set "model": "auto" to let the gateway choose a provider, but naming a Gemini model keeps you on the family you want.

Turning Gemini thinking on when you need it

Some problems deserve deeper reasoning: multi-step math, tricky code, planning, or careful analysis. For those, you opt in per request. There are two ways to enable thinking, and both apply only to Gemini:

• "reasoning_effort": one of none (default, off), low, medium, or high. Higher effort means more thinking tokens and (usually) better answers on hard tasks.
• "thinking": true: a convenience flag that switches thinking on without picking a specific effort level.

Because it is off by default, you never pay the thinking tax unless you ask for it. Here is a Python example using the OpenAI SDK that enables thinking through reasoning_effort in the extra body.

from openai import OpenAI

client = OpenAI(
    base_url="https://dreamprompting.com/api/v1",
    api_key="YOUR_API_KEY",
)

resp = client.chat.completions.create(
    model="google/gemini-2.5-flash",
    messages=[
        {"role": "user", "content": "A train leaves at 2pm going 60mph. Another leaves the same station at 3pm going 80mph on the same track. When does the second catch the first? Show your reasoning."}
    ],
    extra_body={"reasoning_effort": "high"},  # or {"thinking": True}
)

print(resp.choices[0].message.content)

Swap "reasoning_effort": "high" for "thinking": True if you just want thinking on without tuning the level. Want streaming for a chat-style UI? Add "stream": true and read tokens as they arrive.

Fast vs thinking: a worked comparison

Consider the train question above. With the default fast call (thinking off), gemini-3.1-flash-lite returns an answer in a fraction of a second using very few tokens, which is perfect when you trust the model or the task is simple. With "reasoning_effort": "high", Gemini works through the relative speeds, the one-hour head start, and the catch-up time step by step before committing to a final answer. You spend more tokens and a bit more latency, but you get a traceable, more reliable result on the kind of problem where a quick guess often goes wrong. The point of the free Gemini API on DreamPrompting is that this is your choice, request by request, rather than a fixed default someone else picked for you. Full details on thinking control live in the API docs, and you can browse every available model on the models page.

FAQ

Is thinking on by default? No. Gemini defaults to gemini-3.1-flash-lite with thinking off, so you get fast, cheap, predictable answers. Enable it per request with reasoning_effort or "thinking": true.

Which Gemini models are available? The default lite model google/gemini-3.1-flash-lite and the more capable google/gemini-2.5-flash. Call GET /api/v1/models or open the models page for the current list.

Does it cost anything? No. It is a free service running on Google's free AI Studio tier through the gateway, and the API key is free too: sign in with Google, no credit card. Please keep usage within fair use, the 100 requests per minute per IP rate limit, and your account's rolling 24 hour account quota of 5,000 requests and 500,000 tokens.

Start building

You now have everything you need to use the free Gemini API: a default fast path and full thinking control when a problem earns it. Read the API docs for every parameter, or try prompts right now in the chat playground before you write a line of code.