Gemini Decision 1 API
Gemini Decision 1 is a Google decision model that scores up to 1,000 predefined questions per request, returning probabilities and confidence scores instead of generated text.
Gemini Decision 1, released by Google on November 11, 2026, is a decision model available through the Gemini API in Google AI Studio. Rather than generating text, it returns probabilities and confidence scores for predefined questions, scoring up to 1,000 independent questions in a single request in one forward pass. It handles yes/no questions, multiple-choice picks, and rating scores, and returns a confidence score from zero to one for each decision. The model accepts text, image, audio, video, and PDF input with a 256K token context, and supports tuning resolution on image and video inputs to control visual detail and token cost. Developer guides are available in Python, JavaScript, and REST.
The model is built for tasks where a clear decision matters more than prose: tool selection, policy checks, model routing, context grading, content grading, and scanning delivery photos against a checklist of quality criteria. Google reports end-to-end latency of 150 to 250 milliseconds for contexts up to 16K tokens, with latency remaining nearly flat whether 1 or 100 questions are asked per request; shared input material is billed once per request regardless of question count.
On JevBench, Google's own testing shows 92.2% compared with 90.0% for TypeSafe's Jev 1.14. Google also reports the lowest calibration error among compared models and leadership across a suite of benchmarks spanning agent routing, reasoning, ranking, and multimodal understanding. Published latency figures apply to contexts up to 16K tokens rather than the full 256K limit, and independent benchmark verification is not yet available.
Updated:
MODEL ID
gemini-decision-1- Context
- 262.144K
- Max output
- 235.929K
- Input
- text / image / audio / video / file
- Output
- decisions
Highlights
- Scores up to 1,000 questions per request
- Returns probabilities and confidence scores
- 150 to 250 ms latency up to 16K tokens
- 256K token context window
- Text, image, audio, video, and PDF input
- Available in Google AI Studio
Providers
Compare API prices across providers. Select a provider to view all charges and conditions.
| Uptime | |||
|---|---|---|---|
| $0.1 | $0 |
Historical observations are not a live health check. Missing data does not mean an outage.
Benchmarks
- 92.2%
JevBench
Google's own testing; compared with 90.0% for TypeSafe's Jev 1.14
Scores as published by the model provider.
Decisions Quick Start
Send state and narrow typed questions to /api/v1/decisions and read one answer per question: noul returns a probability from 0 to 1, choice the selected option with its distribution, and score a numeric score for ordered criteria. This endpoint is separate from the Chat Completions, Messages, and Responses surfaces.
curl https://openapi.felo.ai/api/v1/decisions \
-H "Authorization: Bearer $FELO_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "gemini-decision-1",
"state": "Help! My payouts have been failing for 3 days.",
"questions": {
"is_urgent": {
"type": "noul",
"instructions": "Does this message convey urgency?",
"criteria": {"true": "Time-sensitive", "false": "No urgency"}
},
"department": {
"type": "choice",
"instructions": "Which team should handle this?",
"criteria": {
"billing": "Payments, invoicing, refunds",
"technical": "Bugs, outages, integrations",
"sales": "Pricing, upgrades, new accounts"
}
},
"frustration": {
"type": "score",
"instructions": "How frustrated is the customer?",
"criteria": ["Calm", "Frustrated", "Very angry"]
}
}
}'Frequently asked questions
What is Gemini Decision 1?
Gemini Decision 1, released by Google on November 11, 2026, is a decision model available through the Gemini API in Google AI Studio. Rather than generating text, it returns probabilities and confidence scores for predefined questions, scoring up to 1,000 independent questions in a single request in one forward pass. It handles yes/no questions, multiple-choice picks, and rating scores, and returns a confidence score from zero to one for each decision. The model accepts text, image, audio, video, and PDF input with a 256K token context, and supports tuning resolution on image and video inputs to control visual detail and token cost. Developer guides are available in Python, JavaScript, and REST. The model is built for tasks where a clear decision matters more than prose: tool selection, policy checks, model routing, context grading, content grading, and scanning delivery photos against a checklist of quality criteria. Google reports end-to-end latency of 150 to 250 milliseconds for contexts up to 16K tokens, with latency remaining nearly flat whether 1 or 100 questions are asked per request; shared input material is billed once per request regardless of question count. On JevBench, Google's own testing shows 92.2% compared with 90.0% for TypeSafe's Jev 1.14. Google also reports the lowest calibration error among compared models and leadership across a suite of benchmarks spanning agent routing, reasoning, ranking, and multimodal understanding. Published latency figures apply to contexts up to 16K tokens rather than the full 256K limit, and independent benchmark verification is not yet available.
How is usage billed?
Your bill depends on the provider you use, its prices and your usage. Longer inputs or other conditions may have different prices.
What inputs does Gemini Decision 1 accept?
It accepts text, image, audio, video, and PDF input, with a 256K token context. Resolution on image and video inputs can be tuned to control visual detail and token cost.
How many questions can it score in one request?
It can score up to 1,000 independent questions in a single request in one forward pass. Supported question types include yes/no questions, multiple-choice picks, and rating scores.
What does the model return instead of text?
It returns probabilities and confidence scores for predefined questions, with a confidence score from zero to one for each decision.
What tasks is it suited for?
It suits tool selection, policy checks, model routing, context grading, content grading, and checking delivery photos against a checklist of quality criteria.
How does it compare with other decision models?
On JevBench, Google's own testing gives it 92.2% versus 90.0% for TypeSafe's Jev 1.14. Google reports the lowest calibration error among compared models, but independent benchmark verification is not yet available.