> ## Documentation Index > Fetch the complete documentation index at: https://evolink.ai/docs/llms.txt > Use this file to discover all available pages before exploring further. # Gemini Decisions - Complete Parameter Reference > - Use `gemini-decision-1` to evaluate text, images, audio, video, and PDFs with yes/no probabilities, single-choice answers, and scores - Returns all answers synchronously; streaming, Batch API, and context caching are not supported - Only input tokens are billed; output is free. A minimum charge applies per request. See [Model pricing](https://evolink.ai/pricing) - For conversation or text generation, use the [Gemini Native API](/en/api-manual/language-series/gemini/native-api/native-api-reference) > [!NOTE] > > **BaseURL**: Use `https://direct.evolink.ai` by default; `https://api.evolink.ai` is the fallback endpoint. ## OpenAPI ````yaml en/api-manual/language-series/gemini/decisions/decisions-reference.json POST /v1beta/decisions openapi: 3.1.0 info: title: Gemini Decisions - Complete Parameter Reference description: >- Use the Gemini Decisions API to have Gemini Decision 1 answer typed questions about text, images, audio, video, and PDF documents: yes/no probabilities, one option from a fixed list, or a rating on ordered levels. license: name: MIT version: 1.0.0 servers: - url: https://direct.evolink.ai description: Production (recommended) - url: https://api.evolink.ai description: Fallback endpoint security: - bearerAuth: [] tags: - name: Decisions description: 'Structured decisions: classification, routing, and scoring' paths: /v1beta/decisions: post: tags: - Decisions summary: Gemini Decision 1 (structured decisions) description: >- - Use `gemini-decision-1` to evaluate text, images, audio, video, and PDFs with yes/no probabilities, single-choice answers, and scores - Returns all answers synchronously; streaming, Batch API, and context caching are not supported - Only input tokens are billed; output is free. A minimum charge applies per request. See [Model pricing](https://evolink.ai/pricing) - For conversation or text generation, use the [Gemini Native API](/en/api-manual/language-series/gemini/native-api/native-api-reference) operationId: createDecision requestBody: required: true content: application/json: schema: $ref: '#/components/schemas/DecisionRequest' examples: noul: summary: Yes/no question value: 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? choice: summary: Single choice (routing) value: model: gemini-decision-1 input: I was charged twice for my order. questions: department: type: choice instructions: Which department should handle this request? criteria: billing: Payments, invoices, and refunds. technical: Problems using the product. shipping: Delivery and tracking. other: Anything outside these categories. score: summary: Scoring on ordered levels value: model: gemini-decision-1 input: Export fails in Safari but works in Chrome. questions: severity: type: score instructions: How severe is this issue for the user? criteria: - 'Cosmetic: appearance only, no lost functionality.' - >- Workaround available: a task fails, but another way works. - 'Fully blocked: a task fails with no workaround.' multiple: summary: Ask multiple questions in one request value: model: gemini-decision-1 input: - type: text text: > {"order": {"id": "8821", "item": "headphones", "price_usd": 240}, "tier": "Gold"} - type: text text: >- {"message": "Audio cuts out on the left earcup. Fix it today or I cancel my membership."} questions: is_defect: type: noul instructions: Does `message` describe a hardware defect? next_action: type: choice instructions: What is the right resolution given `tier` and `order`? criteria: instant_refund: Refund without return. Low value or clear fault. replacement_shipment: Ship a replacement. Customer returns the original. troubleshooting_flow: The issue may be fixable by the customer. human_escalation: '' churn_risk: type: score instructions: >- How likely is the customer to churn if the issue is not resolved now? criteria: - No churn signal - Mild frustration - Mentions alternatives - Threatens to cancel image_base64: summary: Evaluate an inline image (base64) value: model: gemini-decision-1 input: - type: text text: Look at the image. - type: image data: >- iVBORw0KGgoAAAANSUhEUgAAACAAAAAgCAIAAAD8GO2jAAAAKklEQVR42mM4ISdHU8QwasGoBaMWjFowasGoBaMWjFowasGoBaMWDBULACXLED1gHZEpAAAAAElFTkSuQmCC mime_type: image/png resolution: low questions: is_red: type: noul instructions: Is the image mostly red? image_url: summary: Evaluate an image from a URL value: model: gemini-decision-1 input: - type: text text: Product photo. - type: image uri: https://www.gstatic.com/webp/gallery/1.jpg mime_type: image/jpeg resolution: low questions: outdoor: type: noul instructions: Is this photo taken outdoors? quality: type: score instructions: How sharp is the photo? criteria: - blurry - ok - sharp audio_url: summary: Evaluate audio from a URL value: model: gemini-decision-1 input: - type: audio uri: >- https://storage.googleapis.com/generativeai-downloads/data/Apollo-11_Day-01-Highlights-10s.mp3 mime_type: audio/mp3 questions: mentions_launch: type: noul instructions: >- Is the speaker talking about spaceflight or a rocket launch? responses: '200': description: One answer per question, keyed by the question keys in the request content: application/json: schema: $ref: '#/components/schemas/DecisionResponse' example: answers: is_defect: type: noul noul: 0.94 confidence: 0.94 next_action: type: choice choice: human_escalation probabilities: human_escalation: 0.76 replacement_shipment: 0.18 troubleshooting_flow: 0.04 instant_refund: 0.02 confidence: 0.76 churn_risk: type: score score: 2.99 legend: '0': No churn signal '1': Mild frustration '2': Mentions alternatives '3': Threatens to cancel probabilities: '0': 0 '1': 0 '2': 0.01 '3': 0.99 confidence: 0.99 usage: total_input_tokens: 324 input_tokens_by_modality: - modality: text tokens: 324 total_output_tokens: 0 total_tokens: 324 '400': description: >- Invalid request. When the request content is invalid, `code` is `invalid_request` and `message` names the field and the rule that failed. Common causes: an unsupported question `type`, a `choice` or `score` question without `criteria`, fewer than 2 options or levels, an invalid question key, empty `input`, an unknown field (such as `stream`), a `mime_type` that does not match the content, or a `gs://` URI or `files/...` name. Input blocked by safety filters returns `safety`, `prohibited_content`, `blocklist`, `image_safety`, or `content_blocked`. Using this model on another endpoint returns `model_endpoint_mismatch`. HTTP error responses are not billed. Safety feedback in an HTTP 200 response is still billed; see `usage`. content: application/json: schema: $ref: '#/components/schemas/ErrorResponse' examples: invalid_question_type: summary: Unsupported question type value: error: message: >- The value 'bool' is not supported for 'type' at 'questions.q'. Supported values: 'choice', 'noul', 'score'. (request id: 20261112194236591724000xVRbpAJk) type: '' param: '' code: invalid_request safety_blocked: summary: Input blocked by safety filters value: error: message: >- Request blocked due to safety violations (harmful content). Please modify your input and retry. (request id: 20261112194253810462000pRk2HvTe) type: '' param: '' code: safety model_endpoint_mismatch: summary: Model sent to the wrong endpoint value: error: allowed_endpoints: - /v1beta/decisions code: model_endpoint_mismatch expected_endpoint: /v1beta/decisions message: >- The model 'gemini-decision-1' cannot be used with /v1/decisions. Allowed endpoints: /v1beta/decisions. (request id: 20261112194245600264000oIU8uBcu) param: null retryable: false type: invalid_request_error '401': description: 'Unauthorized: API Key is missing or invalid' content: application/json: schema: $ref: '#/components/schemas/ErrorResponse' example: error: code: unauthorized message: >- API key is required (request id: 202611121942512304710006r9sjD0r) param: null type: authentication_error '402': description: Insufficient balance content: application/json: schema: $ref: '#/components/schemas/ErrorResponse' '403': description: >- Permission denied. `permission_denied`: `input` references a Google Files API file, which only the Google project that uploaded it can access; send the media as a public or signed HTTPS URL or as inline base64 instead. This status is also returned when the API Key has a model allowlist that does not include this model content: application/json: schema: $ref: '#/components/schemas/ErrorResponse' examples: permission_denied: summary: Google Files API URI used value: error: message: >- You do not have permission to access the File abc123xyz or it may not exist. (request id: 20261112194238873347000q1Tkqeid) type: '' param: '' code: permission_denied '404': description: >- Not found. `not_found`: a media URL in `input` could not be fetched; make sure it is a publicly reachable or correctly signed HTTPS URL. `model_not_found`: the model is not available for the Decisions API; currently only `gemini-decision-1` is supported content: application/json: schema: $ref: '#/components/schemas/ErrorResponse' examples: url_not_found: summary: Media URL could not be fetched value: error: message: >- Cannot fetch content from the provided URL. The URL could not be found. Please double-check the URL for any errors. (request id: 20261112194241224738000Vx0uduIQ) type: '' param: '' code: not_found model_not_found: summary: Model not available for Decisions value: error: code: model_not_found message: >- Model 'gemini-3.8-flash' is not available for the Decisions API (POST /v1beta/decisions) with this API key. This error is permanent — do not retry with the same model id. Call GET /v1/models: Decisions models for this endpoint are listed as gemini-decision-*. (request id: 20261112194249376584000tUELY8LN) param: null retryable: false type: invalid_request_error '429': description: Rate limit exceeded; retry later content: application/json: schema: $ref: '#/components/schemas/ErrorResponse' '500': description: Internal server error content: application/json: schema: $ref: '#/components/schemas/ErrorResponse' '503': description: Service temporarily unavailable; retry later content: application/json: schema: $ref: '#/components/schemas/ErrorResponse' components: schemas: DecisionRequest: type: object additionalProperties: false required: - model - input - questions properties: model: type: string description: >- Model ID. Use `gemini-decision-1`. This model is not a chat model and only works on this endpoint. For example, sending it to the GPT Decisions endpoint `/v1/decisions` returns `400 model_endpoint_mismatch`. enum: - gemini-decision-1 example: gemini-decision-1 input: description: >- Shared input for all questions: a text string, a single content block, or an array of content blocks (text, image, audio, video, or PDF document). Input cannot be empty. Adjacent text blocks are joined in order with no separator, so add your own spaces or newlines. To pass JSON data, serialize it to a string and send it as text. oneOf: - type: string minLength: 1 example: I was charged twice for my order. - $ref: '#/components/schemas/InputContent' - type: array minItems: 1 items: $ref: '#/components/schemas/InputContent' questions: type: object description: >- Questions to evaluate, as a map from question key to question: 1 to 1,000 questions. You choose the keys (1 to 64 characters matching `[A-Za-z0-9_-]+`), and each answer is returned under the same key. The model never sees the keys, so write the full question in `instructions`. | Type | Purpose | Main result | |---|---|---| | `noul` | Check whether a statement holds | `noul`: probability of “yes” (0.01 to 0.99) | | `choice` | Pick one option from a fixed list | `choice`: the most likely option ID | | `score` | Rate the input on ordered levels | `score`: probability-weighted level score | Each question is evaluated independently against the same input and cannot see the others; question types can be mixed. If questions depend on one another, split them into separate requests. The input and all questions share a 262,144-token context window. minProperties: 1 maxProperties: 1000 propertyNames: pattern: ^[A-Za-z0-9_-]{1,64}$ additionalProperties: oneOf: - $ref: '#/components/schemas/NoulQuestion' - $ref: '#/components/schemas/ChoiceQuestion' - $ref: '#/components/schemas/ScoreQuestion' example: is_damaged: type: noul instructions: Did the item arrive damaged? description: >- The request body accepts only `model`, `input`, and `questions`. Any other field (such as `stream`) returns `400 invalid_request`. DecisionResponse: type: object properties: answers: type: object description: >- One answer per question, keyed by the question keys in the request. Each answer’s `type` matches its question. Empty when `prompt_feedback.block_reason` is set additionalProperties: oneOf: - $ref: '#/components/schemas/NoulAnswer' - $ref: '#/components/schemas/ChoiceAnswer' - $ref: '#/components/schemas/ScoreAnswer' usage: $ref: '#/components/schemas/DecisionUsage' prompt_feedback: $ref: '#/components/schemas/PromptFeedback' ErrorResponse: type: object properties: error: type: object properties: message: type: string description: >- Error description, ending with `(request id: ...)`. Provide this ID when troubleshooting type: type: - string - 'null' description: Error type; may be empty param: type: - string - 'null' description: Related parameter; may be empty code: type: - string - 'null' description: >- Error code, such as `invalid_request`, `permission_denied`, `not_found`, `safety`, `model_not_found`, or `model_endpoint_mismatch` InputContent: description: A content block. `type` determines the other fields oneOf: - $ref: '#/components/schemas/TextContent' - $ref: '#/components/schemas/ImageContent' - $ref: '#/components/schemas/AudioContent' - $ref: '#/components/schemas/VideoContent' - $ref: '#/components/schemas/DocumentContent' NoulQuestion: type: object description: Yes/no question required: - type - instructions properties: type: type: string enum: - noul instructions: $ref: '#/components/schemas/QuestionInstructions' criteria: type: object description: >- Optional. Defines what “yes” and “no” mean when the question alone is ambiguous properties: true_description: type: string description: When the answer is “yes” false_description: type: string description: When the answer is “no” ChoiceQuestion: type: object description: Single-choice question required: - type - instructions - criteria properties: type: type: string enum: - choice instructions: $ref: '#/components/schemas/QuestionInstructions' criteria: type: object description: >- Options, as a map from option ID to a description of when it applies (use an empty string `""` when the ID is self-explanatory). 2 to 255 options. Option IDs are 1 to 128 characters (up to 512 UTF-8 bytes); internal ASCII spaces are allowed, but leading or trailing spaces and control characters are not. Option IDs are sorted alphabetically (UTF-8 byte order) before evaluation, and exact ties resolve to the alphabetically first ID, so use a non-positional catch-all ID such as `other` rather than `none_of_the_above` minProperties: 2 maxProperties: 255 additionalProperties: type: string example: billing: Payments, invoices, and refunds. technical: Problems using the product. shipping: Delivery and tracking. other: Anything outside these categories. ScoreQuestion: type: object description: Score question required: - type - instructions - criteria properties: type: type: string enum: - score instructions: $ref: '#/components/schemas/QuestionInstructions' criteria: type: array description: >- Level descriptions ordered from lowest to highest: 2 to 255 non-empty strings (2 to 10 recommended). Level `i` (starting at 0) is described by `criteria[i]`. Because `score` averages level indices, keep the scale linear and adjacent levels clearly distinct minItems: 2 maxItems: 255 items: type: string minLength: 1 example: - 'Cosmetic: appearance only, no lost functionality.' - 'Workaround available: a task fails, but another way works.' - 'Fully blocked: a task fails with no workaround.' NoulAnswer: type: object properties: type: type: string enum: - noul noul: type: number minimum: 0.01 maximum: 0.99 description: >- Calibrated probability that the answer is “yes”, from 0.01 to 0.99; the probability of “no” is `1 - noul`. Set thresholds within this range using labeled examples from your own traffic confidence: type: number minimum: 0 maximum: 1 description: >- Estimated probability that the top answer is right, equal to `max(noul, 1 - noul)` ChoiceAnswer: type: object properties: type: type: string enum: - choice choice: type: string description: The option ID with the highest probability probabilities: type: object description: >- Calibrated probability for each option ID, summing to 1. Probability is always spread across your options, even when none fits additionalProperties: type: number confidence: type: number minimum: 0 maximum: 1 description: >- Probability of the selected option, from 0 to 1. When it is low, consider human review or a separate yes/no scope check ScoreAnswer: type: object properties: type: type: string enum: - score score: type: number description: >- Probability-weighted average of the level indices (`sum(k × probability_k)`), so it can fall between levels. Divide it by the highest level index to normalize it to 0–1 probabilities: type: object description: >- Calibrated probability for each level index (`"0"`, `"1"`, ...), summing to 1. For a discrete level, take the index with the highest probability additionalProperties: type: number legend: type: object description: Description of each level index, taken from `criteria` additionalProperties: type: string confidence: type: number minimum: 0 maximum: 1 description: The highest probability in `probabilities`, from 0 to 1 DecisionUsage: type: object description: >- Token usage. Only input tokens are billed (`total_input_tokens`): text, image, audio, video, and PDF tokens cost the same, and output is always 0. A minimum charge applies to each request. See [Model pricing](https://evolink.ai/pricing) for current prices. Context caching is not supported; repeated input is billed separately for each request. If an HTTP 200 response contains `prompt_feedback.block_reason`, input usage is still billed. If valid usage is not returned, billing uses an estimate. properties: total_input_tokens: type: integer description: >- Billable input tokens for the shared input and all questions. The shared input counts once per request regardless of how many questions you ask; each question adds only its own instructions and criteria example: 324 input_tokens_by_modality: type: array description: Input tokens by modality items: type: object properties: modality: type: string description: >- Modality: `text`, `image`, or `audio`. Video frames and PDF page images are counted as `image` tokens: type: integer description: Token count for this modality total_output_tokens: type: integer description: Always 0; this model generates no text example: 0 total_tokens: type: integer description: Total tokens example: 324 PromptFeedback: type: object description: >- Feedback from input content filters. When `block_reason` is set, the input was blocked, no questions were evaluated, and `answers` is empty. If returned with HTTP 200, the request is still billed; see `usage`. properties: block_reason: type: string enum: - safety - other - blocklist - prohibited_content - image_safety description: Why the input was blocked block_reason_message: type: string description: Details about why the input was blocked TextContent: type: object required: - type - text properties: type: type: string enum: - text text: type: string description: Text content ImageContent: type: object required: - type - mime_type properties: type: type: string enum: - image data: type: string contentEncoding: base64 description: >- Base64-encoded file bytes (inline data is limited to 100 MB per request in total). Provide either `data` or `uri` uri: type: string format: uri description: >- Public or signed HTTPS URL of the file (up to 15 MB). Google Files API URIs, `gs://` URIs, `files/...` names, and YouTube URLs are not supported. Provide either `data` or `uri` mime_type: type: string enum: - image/png - image/jpeg - image/webp - image/heic - image/heif - image/gif - image/bmp - image/tiff description: >- MIME type. It must match the real file format; generic types such as `application/octet-stream` are rejected resolution: type: string enum: - low - medium - high - ultra_high description: >- Optional. Media resolution, which controls token usage and visual detail AudioContent: type: object required: - type - mime_type properties: type: type: string enum: - audio data: type: string contentEncoding: base64 description: >- Base64-encoded file bytes (inline data is limited to 100 MB per request in total). Provide either `data` or `uri` uri: type: string format: uri description: >- Public or signed HTTPS URL of the file (up to 15 MB). Google Files API URIs, `gs://` URIs, `files/...` names, and YouTube URLs are not supported. Provide either `data` or `uri` mime_type: type: string enum: - audio/wav - audio/mp3 - audio/mpeg - audio/aiff - audio/aac - audio/ogg - audio/flac - audio/m4a - audio/opus - audio/webm - audio/l16 - audio/alaw - audio/mulaw description: >- MIME type. It must match the real file format; generic types such as `application/octet-stream` are rejected VideoContent: type: object required: - type - mime_type properties: type: type: string enum: - video data: type: string contentEncoding: base64 description: >- Base64-encoded file bytes (inline data is limited to 100 MB per request in total). Provide either `data` or `uri` uri: type: string format: uri description: >- Public or signed HTTPS URL of the file (up to 15 MB). Google Files API URIs, `gs://` URIs, `files/...` names, and YouTube URLs are not supported. Provide either `data` or `uri` mime_type: type: string enum: - video/mp4 - video/mpeg - video/mov - video/avi - video/webm - video/wmv - video/x-flv - video/3gpp description: >- MIME type. It must match the real file format; generic types such as `application/octet-stream` are rejected resolution: type: string enum: - low - medium - high description: >- Optional. Media resolution, which controls token usage and visual detail. `ultra_high` applies to images only DocumentContent: type: object required: - type - mime_type properties: type: type: string enum: - document data: type: string contentEncoding: base64 description: >- Base64-encoded PDF bytes (up to 50 MB). Provide either `data` or `uri` uri: type: string format: uri description: >- Public or signed HTTPS URL of the file (up to 15 MB). Google Files API URIs, `gs://` URIs, `files/...` names, and YouTube URLs are not supported. Provide either `data` or `uri` mime_type: type: string enum: - application/pdf description: >- MIME type. It must match the real file format; generic types such as `application/octet-stream` are rejected QuestionInstructions: description: >- The question to evaluate against `input`, written in full. Text only: a string, a text block, or an array of text blocks oneOf: - type: string minLength: 1 - $ref: '#/components/schemas/TextContent' - type: array minItems: 1 items: $ref: '#/components/schemas/TextContent' securitySchemes: bearerAuth: type: http scheme: bearer description: |- Authenticate with your EvoLink API Key as a Bearer token. **Get an API Key:** Visit the [API Key management page](https://evolink.ai/dashboard/keys). **Request header:** ``` Authorization: Bearer YOUR_API_KEY ``` ```` This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.