Gemini Decision 1 Pricing: $0.10/M Input, API, and Use Cases

Gemini Decision 1 is for the moment when your program needs a judgment, not a paragraph. It answers yes/no questions, selects from your options, or scores against a rubric, returning probabilities instead of generating answer text. Google's official price is $0.10 per million input tokens, with zero output tokens.

The practical trade-off is straightforward: use it for bounded decisions that your code can act on. Keep a generative model when the job needs an explanation, extracted free-form values, tool arguments, or multi-step reasoning. AIHubMix lists Gemini Decision 1 on a separate decision endpoint; it is not a model to drop into Chat Completions.

Three answers your code can work with

Released on November 11, 2026, Google's specialized decision model evaluates questions independently over shared input. Google's native API accepts text, images, audio, video, and PDFs, within a 262,144-token window covering the input and all questions.

Type	Your question	Answer value
noul	Does this condition hold?	Probability of yes
choice	Which route fits?	One option plus probabilities
score	How severe is this?	Expected rubric level

For the native Google API, noul ranges from 0.01 to 0.99. A choice has 2–255 options. A score has 2–255 ordered levels, although Google recommends 2–10 for most rubrics. Its result can be fractional: the score is the probability-weighted average of level indices, not necessarily the most likely level.

Google allows up to 1,000 questions per request. That is useful for independent checks, not a hidden reasoning chain: one question cannot read another question's answer. Combine outcomes in code, or make another request when a later decision genuinely depends on an earlier one.

These are native Google capabilities. They do not establish that every media type, limit, or response field is available through an intermediary.

Pricing: count the whole request, not just the ticket

The official pricing page bills shared input once per request. Each question adds its instructions, criteria, and formatting tokens. Three questions about one ticket do not mean three separately billed copies of that ticket, but they are not token-free either.

Model	Input / 1M tokens	Output / 1M tokens
Gemini Decision 1	$0.10	No output tokens
Gemini 3.5 Flash-Lite	$0.30	$2.50

These are official standard direct-API prices in USD. Flash-Lite's output price includes thinking tokens. AIHubMix's listed rates for these two models match those prices at the time of checking, November 12, 2026.

Consider an illustrative, not measured, workload: 100,000 tickets per month, each sent once with three questions, using 2,000 billed input tokens per request including the questions.

Decision: 200 million input tokens × $0.10/M = $20.
Gemini 3.5 Flash-Lite: the same assumed input costs $60. Assume another 40 output tokens per request: four million output tokens cost $10, for $70 total.

This excludes caching, retries, taxes, and any downstream review. It also assumes the same billed input count for the illustration, not equal tokenization or equal quality. If Decision makes the wrong routing call more often on your data, the lower token bill may not be the lower operating cost.

The Decisions API has no context caching or Batch API. Repeating a long state in the next request bills it again; the shared-input benefit is within a request, not across requests.

The API boundary matters

The AIHubMix model instructions document POST /v1/systemone, Bearer authentication, and a body containing model, state, and questions.

Google's native Decisions API instead uses /v1beta/decisions and an input field. They describe the same kind of task, but the request envelopes are not interchangeable. Do not point Google's native SDK example at another base URL and assume the method and fields translate automatically.

The following is a documentation-based text example, not an executed test. It uses a fictional ticket and deliberately avoids media inputs whose gateway behavior has not been verified here:

curl --fail-with-body https://aihubmix.com/v1/systemone \
  -H "Authorization: Bearer $AIHUBMIX_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gemini-decision-1",
    "state": "The export button fails in Safari. Chrome works, but I need to send the report today.",
    "questions": {
      "department": {
        "type": "choice",
        "instructions": "Which team should handle this support request?",
        "criteria": {
          "billing": "Charges, invoices, or refunds.",
          "technical": "A product feature is not working.",
          "other": "The request does not fit either department."
        }
      },
      "urgency": {
        "type": "score",
        "instructions": "How time-sensitive is the request?",
        "criteria": [
          "No deadline is stated.",
          "A future deadline is stated, but not today.",
          "The user needs the task completed today."
        ]
      },
      "has_workaround": {
        "type": "noul",
        "instructions": "Does the message explicitly describe a working alternative?"
      }
    }
  }'

choice.criteria is an object; score.criteria is an ordered array. Keep them distinct. Put the complete question in instructions, rather than relying on the question key to convey its meaning.

One integration caveat deserves attention: the current AIHubMix instructions include a response example naming a Jev model. That example does not establish Gemini's actual response shape, routing, output-token accounting, or score and confidence semantics. This article has not made an authenticated request; validate the real returned types and usage before integration, rather than treating that example as a Gemini transcript. Full native media support, context limits, and question-count limits through this channel also remain unverified here.

Turn probabilities into a policy, not an unchecked action

The Google guide recommends setting thresholds against labeled examples from your own traffic. For a routing workflow:

Validate the response and required fields before reading an answer.
Check that a selected route belongs to your allowed options.
Send uncertain, out-of-scope, or malformed results to review.
Keep a separate held-out set to measure errors after selecting thresholds.

On Google's native API, confidence equals the highest outcome probability; for a yes/no answer it is max(p, 1 − p). It is not a second model independently confirming the decision. A high-confidence “no” is still a no, so do not substitute confidence for the probability that a condition holds.

For scores, decide whether your policy needs the expected score or the single most likely level. Those can produce different actions. Include an other option for choices that may be out of scope; a closed option list still distributes probability when no option fits.

There is another boundary for safety filters: Google's documentation shows blocked inputs returning HTTP 400 rather than decision answers. That is a native documentation example, not a measured AIHubMix response. Handle non-success responses separately, using the gateway's HTTP error guidance, and never turn a missing answer into automatic approval.

When it fits—and when it does not

Decision is a plausible fit for request routing, policy checks, candidate ranking, and narrow agent stop/escalation decisions. Google reports median end-to-end latency of 150–250 ms for inputs up to 16K tokens and nearly flat latency from 1 to 100 questions. Those are vendor-reported figures, not AIHubMix measurements or a promise for 1,000 questions or the full context window.

Stay with a generative model when you need new text, arbitrary extracted values, or tool calls with arguments. A typed answer also does not grant permission to execute an action: authorization and high-stakes review remain application responsibilities.

FAQ

What does Gemini Decision 1 cost?
Google's standard price is $0.10 per million input tokens. It generates zero output tokens. Question instructions and criteria contribute to input usage.

Is it a Chat Completions model?
No. AIHubMix documents a System One decision endpoint. Google's native service uses the Decisions API with a different request envelope.

Does zero output tokens mean an empty response?
No. The API returns typed answers and probabilities; it does not generate answer text token by token.

Can one request contain 1,000 questions?
Google's native API allows up to 1,000 within its total context limit. This article has not verified that maximum through AIHubMix.

Can it process images, audio, video, and PDFs?
Google's native API documents all four alongside text. Their availability through AIHubMix has not been verified in this article; the example uses text only.

Is the example above tested?
No. It follows the gateway's published request format. No authenticated request, measured output, or latency test was run for this article.

Keep reading: the Gemini Decision 1 series

Choosing a decision model rather than just an endpoint? Compare calibration, tool routing, and cost in Gemini Decision 1 vs Jev 1.14: Which Should You Choose?.

Sources

Official and gateway prices rechecked November 12, 2026, at 06:43 UTC. Specifications and prices are documentation snapshots, not inference measurements.