Gemini Decision: How to Use It Without a Google Account or Credit Card

Q: Can I call it on Moyi API?

Not yet. It runs on /v1beta/decisions, which our gateway doesn't forward. It will appear on the model and pricing pages once we support it.

Short answer

Google released Gemini Decision in the Gemini API on November 11, 2026. The model ID is gemini-decision-1. One request asks up to 1,000 yes/no, single-choice or rubric questions about the same input — text, images, audio, video or PDF — and returns calibrated probabilities. It generates zero output tokens, so you only pay for input: $0.10 per million input tokens, with a free tier.

There is one catch right now: it lives on its own endpoint, POST /v1beta/decisions, not the usual generateContent. You can't just swap the model ID in existing code — the client has to speak the Decisions API request shape.

To use it today: a Google account for a Gemini API key, plus a network that can reach generativelanguage.googleapis.com.
Moyi API does not list gemini-decision-1 yet. Our gateway forwards the /v1beta/models/{model}:generateContent family; /v1beta/decisions isn't in the routing table. Once we add it, a Moyi API key will work the same way as for other models.

How it differs from a normal Gemini model

	Gemini Decision	Chat models like Gemini 4 Carbon
Endpoint	`POST /v1beta/decisions`	`POST /v1beta/models/{model}:generateContent`
Questions per request	1 to 1,000	one turn
Returns	probabilities, confidence, answers keyed by question	generated text
Output tokens	zero, not billed	billed by output
Price per 1M input tokens	$0.10	$2 for Carbon (introductory)
Input limit per request	262,144 tokens	model context window

What it's for: reading one piece of material once and deciding many things at the same time. A support message that needs "is it damaged?", "which queue?", and "how severe?" answered in one call; or scoring a batch of candidates against a rubric. The value is one prefill shared across many questions — not text generation.

What it can't do: no Batch API, no context caching, no Flex or Priority inference, and it isn't a general chat model. Identical input across separate requests is billed in full each time. Free-tier limits: 200 requests per minute, 50,000 tokens per minute, 10,000 requests per day.

Sources: Google's Gemini API release notes, the Gemini Decision model page and the Decisions API guide (November 11, 2026).

What you can do today, by region

Short version: only the official route works right now, which means a Google account, a network that reaches Google, and a payment method Google accepts.

Through Google directly

Get an API key from Google AI Studio with a Google account.
Your network must reach generativelanguage.googleapis.com. From mainland China this domain is not reachable directly — you need a network environment that can reach Google. There is no workaround at the proxy layer for this, because the endpoint itself lives on Google's servers.
The free tier needs no card: 200 requests/min, 50,000 tokens/min, 10,000 requests/day. Going past that means the paid tier and a payment method Google accepts.

In Russia: Google's payment methods are restricted there, so adding a card may not go through. The free tier works without a card for evaluating the model, but it isn't something to run production traffic on.

Through a third-party API gateway

The usual pitch for these gateways is "no Google account, local payment methods, one key for many models". Gemini Decision is an exception right now. Because it isn't on the standard generateContent path, a gateway has to add a dedicated /v1beta/decisions forward and handle its specific request shape (model, input, questions) and response shape.

Moyi API hasn't added that route yet, so calling gemini-decision-1 with a Moyi API key will not work today. When we do add it, it will show up on the pricing page, and the same key will work with just a model-ID change.

If you just need classification in production now

Gemini Decision's selling point is many questions per call at zero output cost. You can get a workable version today with a chat model — more expensive per call, and you write the prompt so the output is structured:

Gemini 4 Carbon: the strongest Gemini 4 reasoning model, $2 / $10 introductory through November 30.
Gemini 4 Argon: the first Gemini 4 model, $2 / $10.

Both work through a Moyi API key in the OpenAI-compatible format, with no Google account. See Gemini 4 Argon API.

The official request format

The REST form from Google's docs:

bash

curl -X POST "https://generativelanguage.googleapis.com/v1beta/decisions" \
  -H "x-goog-api-key: $GEMINI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gemini-decision-1",
    "input": "The blue mug I ordered on Monday arrived cracked on Wednesday.",
    "questions": {
      "is_damaged": { "type": "noul", "instructions": "Did the item arrive damaged?" },
      "queue": {
        "type": "choice",
        "instructions": "Which support queue should handle this message?",
        "criteria": {
          "billing": "Payment, invoice, or refund questions",
          "replacements": "Damaged, missing, or wrong items",
          "general": "Product and account questions"
        }
      }
    }
  }'

Two details that trip people up:

The auth header is x-goog-api-key, not Authorization: Bearer.
You choose the question keys (is_damaged, queue here) and answers come back under the same keys.

With the Python SDK it's client.decisions.create(...), and api_version must be v1beta:

python

from google import genai

client = genai.Client(http_options={"api_version": "v1beta"})
result = client.decisions.create(
    model="gemini-decision-1",
    input="The blue mug I ordered on Monday arrived cracked on Wednesday.",
    questions={
        "is_damaged": {"type": "noul", "instructions": "Did the item arrive damaged?"},
    },
)

FAQ

Is Gemini Decision released? Yes. Google shipped it in the Gemini API on November 11, 2026, model ID gemini-decision-1.

Can I call it on Moyi API? Not yet. It runs on /v1beta/decisions, which our gateway doesn't forward. It will appear on the model and pricing pages once we support it.

Why can't I just change the model name like with other Gemini models? Because it doesn't generate text — it asks many questions and returns probabilities. Both the request shape (questions) and the response shape (per-question probabilities and confidence) differ, and the standard generateContent endpoint can't carry them.

How much does it cost? $0.10 per million input tokens; output tokens are zero and not billed. Free tier: 200 RPM, 50,000 TPM, 10,000 RPD.

Does it work in China? Only through Google today, which needs a network that reaches generativelanguage.googleapis.com. Direct connections from mainland China don't work.

Does it work in Russia? The free tier works without a card for testing; the paid tier needs a payment method Google accepts, which is restricted there.

How many questions fit in one request? Up to 1,000, bounded by the 262,144-token context window. Choice and rubric questions support 2 to 255 options or levels.

Keep reading

All posts