> ## Documentation Index > Fetch the complete documentation index at: /docs/llms.txt > Use this file to discover all available pages before exploring further. # Criar decisões > **Beta.** Evaluate content against named questions and receive structured answers, probabilities, and usage costs. Send the content as `state` and define each entry in `questions` as: - `noul`: estimate the probability that a statement is true. - `choice`: select an option from a set. - `score`: rate the content on an ordered scale. - `box`: locate one target in an image and return its bounding box. Beta, TypeSafe `jev-preview` only. Instructions and criteria accept strings or structured JSON. TypeSafe models also accept inline images through `assets` (beta). Use a native decision model or a supported chat model. Configure ordered `fallbacks`, optional `retry`, `timeout.call_timeout` in milliseconds, and an exact-match `cache`. Each retry and fallback gets a fresh timeout; when `retry` is omitted, the gateway retries twice before moving to the next fallback. The response identifies the model that answered. Requires the `classify` API-key permission. PII plugins and guardrails are not applied. See the [Decisions guide](/ai-gateway/features/decisions) for supported models, probability interpretation, and refusals. Consulte [URLs e autenticação HTTP](/docs/pt-BR/reference/base-urls.md) e a referência da API para os formatos completos de solicitação e resposta. > [!NOTE] > > **Guia relacionado**: consulte o [guia de Decisions](/docs/pt-BR/ai-gateway/features/decisions.md) para exemplos, fallbacks, timeouts, retries, cache e atribuição de identidade. A requisição exige a permissão de chave de API `classify`. ## OpenAPI ````yaml post /v3/router/decisions openapi: 3.1.0 info: title: Toqan Router API version: '2.0' description: Toqan Router API documentation servers: - url: https://router.toqan.ai/eu description: Frankfurt (Europe) - url: https://router.toqan.ai/br description: São Paulo (Brazil) security: - ApiKey: [] tags: - description: >- API keys authenticate programmatic access to the workspace. They expose opaque tokens, per-domain access grants, and budget and rate-limit constraints. name: API keys - name: Logs description: >- OpenTelemetry log query API. Search, filter, aggregate, and facet log records ingested via OTLP. - name: Reporting description: >- GenAI reporting API over canonical analytics rollups. Accepts a metric name, time range, grain, group-by, and filters; returns a typed time series and optional totals. - name: Telemetry description: >- Unified query envelope for traces, metrics, and logs. One request shape, one filter dialect, and one response shape per source, validated by a per-source registry. - name: Traces description: >- Query and inspect ingested trace data: search trace summaries, aggregate metrics, and read individual traces and their spans. - name: Views description: Saved filters and columns for workspace observability pages. - description: List models available through the AI Router. name: Models - name: Alerts description: >- Alerts evaluate a telemetry metric on a fixed interval and fire notifications through notifiers when the value breaches a threshold. Each breach opens a trigger that tracks the incident until the value recovers. - name: Annotation Queues description: Annotation queues collect spans for human review. - name: Budgets description: >- Budgets govern spend, token usage, and request rate across six scopes: workspace, project, identity, API key, provider, and model. Every applicable budget is enforced, and the most restrictive limit applies per dimension. - name: Files description: File upload and retrieval operations. - name: Guardrail Rules description: >- Guardrail Rules conditionally enforce evaluators and plugins for AI Gateway traffic. Rules may be scoped to a project or the whole workspace. - name: Identities description: >- Identities represent end users from your system for usage and engagement tracking. - name: Management keys description: >- Management keys are workspace-scoped credentials that authenticate programmatic access to workspace administration surfaces (API keys, budgets). Unlike project-scoped API keys, a management key always operates at the workspace level. - name: MCP Gateway description: >- Register upstream MCP servers, discover and sync their tools, and assemble gateways that expose a curated tool surface to MCP clients. - name: Model Catalog description: >- Browse the Toqan Router model catalog: every model Toqan Router offers, across every provider, with pricing, capabilities and benchmark data. List endpoints only return models that are not deprecated. This API is public, requires no authentication, and is rate limited to 120 requests per minute per IP. Responses carry a 5-minute cache-control max-age. - name: Monitors description: >- Monitors are custom observability dashboards for traces, runtime metrics, and logs. Each monitor holds widgets that chart one telemetry query. List the metrics, operations, and dimensions available to your workspace with `GET /v3/telemetry/capabilities`, or start from a preset with `GET /v2/monitors/presets`. - name: Notifiers description: Notifier destinations used to send delivery and workflow notifications. - name: Routing Rules description: >- Routing Rules conditionally select models and enforce request plugins for AI Gateway traffic. Rules are evaluated by ascending priority and may be scoped to a project or the whole workspace. - name: Threads description: Threads group related trace invocations and their aggregate usage - name: Smart Routers description: >- Create and manage workspace Smart Routers. A Smart Router selects a model from an eligible pool for each request according to a quality, balanced, or cost profile. - name: Responses - name: Embeddings - description: Beta. Run typed classification questions against a classify model. name: Classify - description: Beta. Run classification decisions with the classify contract. name: Decisions - description: >- Run large offline workloads as JSONL batches and download OpenAI-compatible result files. name: Batches paths: /v3/router/decisions: post: tags: - Decisions summary: Decisions description: >- **Beta.** Evaluate content against named questions and receive structured answers, probabilities, and usage costs. Send the content as `state` and define each entry in `questions` as: - `noul`: estimate the probability that a statement is true. - `choice`: select an option from a set. - `score`: rate the content on an ordered scale. - `box`: locate one target in an image and return its bounding box. Beta, TypeSafe `jev-preview` only. Instructions and criteria accept strings or structured JSON. TypeSafe models also accept inline images through `assets` (beta). Use a native decision model or a supported chat model. Configure ordered `fallbacks`, optional `retry`, `timeout.call_timeout` in milliseconds, and an exact-match `cache`. Each retry and fallback gets a fresh timeout; when `retry` is omitted, the gateway retries twice before moving to the next fallback. The response identifies the model that answered. Requires the `classify` API-key permission. PII plugins and guardrails are not applied. See the [Decisions guide](/ai-gateway/features/decisions) for supported models, probability interpretation, and refusals. operationId: CreateDecisions requestBody: content: application/json: examples: fallback_retry_identity: summary: Decisions with fallbacks, retries and identity value: fallbacks: - model: openai/gpt-5.6-luna identity: id: customer-demo display_name: Sample customer model: openai/gpt-6-luna questions: positive: criteria: 'false': The customer is unhappy. 'true': The customer is happy. instructions: Is the sentiment positive? type: noul rating: criteria: - Negative - Neutral - Positive instructions: Rate sentiment. type: score sentiment: criteria: negative: Negative sentiment neutral: null positive: Positive sentiment instructions: Classify sentiment. type: choice retry: count: 2 on_codes: - 429 - 502 - 503 - 504 state: 'The customer says: I love this product. It is wonderful!' timeout: call_timeout: 2000 openai_inline_image: description: >- A small self-contained PNG. Replace it with your own base64 image data URL. summary: Classify an inline image with OpenAI Decisions value: model: openai/gpt-6-luna questions: contains_text: instructions: Does the image contain text? type: noul state: - content: - text: Evaluate this image. type: input_text - image_url: >- data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAoAAAAKCAYAAACNMs+9AAAACXBIWXMAAAsTAAALEwEAmpwYAAAAAXNSR0IArs4c6QAAAARnQU1BAACxjwv8YQUAAACLSURBVHgBdY8NDYAgEIXBBEQgAhG0gRGMYBNtoA2M4EygDYygDfCxvdMbk7d9A453f8ZQMcYAWuBVzIHujeGyxE8nYz7dJWYl01p749zxDKABEzhYfDNaMPZSAUymJLZLuvSsSVUhx4G7VM2xpSzQ5YYhLUHDCGoad47ixXjxY84qi1a9QPgZpdYLPVkbtsfywz3jAAAAAElFTkSuQmCC type: input_image role: user openai_text: summary: Classify text with OpenAI Decisions value: model: openai/gpt-6-luna questions: positive: criteria: 'false': The customer is unhappy. 'true': The customer is happy. instructions: Is the sentiment positive? type: noul rating: criteria: - Negative - Neutral - Positive instructions: Rate sentiment. type: score sentiment: criteria: negative: Negative sentiment neutral: null positive: Positive sentiment instructions: Classify sentiment. type: choice state: 'The customer says: I love this product. It is wonderful!' support_ticket: summary: Classify a support message value: model: typesafe/jev-latest questions: is_complaint: instructions: Is the customer complaining? type: noul severity: criteria: - Minor - Moderate - Severe instructions: How severe is the issue? type: score topic: criteria: delivery: Shipping or delivery issues other: null product: Product quality instructions: What is the message mainly about? type: choice state: The parcel arrived two days late and the box was crushed. typesafe_image_box: description: >- Beta. Replace data_base64 with your own PNG or JPEG of 256 to 1568 px per side. summary: Locate an object in an image with TypeSafe value: assets: kitchen-photo: type: image data_base64: format: jpg model: typesafe/jev-preview questions: mug: instructions: the coffee mug in `photo` type: box mug_visible: instructions: Is there a coffee mug in `photo`? type: noul state: photo: kitchen-photo schema: additionalProperties: false properties: assets: additionalProperties: $ref: '#/components/schemas/ClassifyAsset' description: >- Beta, TypeSafe models only. Inline images keyed by a reference string. Write the reference in state, instructions or criteria wherever the image belongs; every asset must be referenced. Required for box questions. type: object cache: $ref: '#/components/schemas/ClassifyCacheConfig' description: >- Exact-match response cache for this request. An identical request body from the same workspace and project is served from the stored response, with the X-Cache header reporting HIT or MISS. fallback_on_codes: description: >- HTTP status codes that trigger fallback. Defaults to [408,429,500,502,503,504,529]. Each code must be unique and between 400 and 599. items: format: int64 type: integer maxItems: 20 type: - array - 'null' fallbacks: description: >- Up to 10 fallback models in order. The gateway retries a model for matching error codes before trying the next model. Every model must support classification and satisfy access checks. Requests with images in state require native OpenAI image-capable fallbacks; requests with assets or box questions require TypeSafe fallbacks. items: $ref: '#/components/schemas/FallbackConfig' maxItems: 10 type: - array - 'null' identity: $ref: '#/components/schemas/ResponseIdentity' description: The identity to attach to the trace. metadata: additionalProperties: type: string description: Key-value metadata attached to the trace. type: object model: description: >- ID of a model that supports native or emulated classify, including openai/gpt-6-luna. type: string name: type: string description: >- The name to display on the trace. If not specified, the default system name will be used. prompt_cache_key: description: >- Native OpenAI only. Partitions reuse of shared input across requests. maxLength: 64 type: string prompt_cache_options: $ref: '#/components/schemas/ClassifyPromptCacheOptions' description: >- Native OpenAI prompt caching controls. Separate from the gateway exact-match response cache. questions: additionalProperties: oneOf: - description: >- Answers with a probability between 0 and 1 that the statement holds. properties: criteria: description: Optional descriptions of what true and false mean. properties: 'false': anyOf: - title: String type: string - additionalProperties: true title: Object type: object - items: {} title: Array type: array description: >- What false means. A string, an object or an array. 'true': anyOf: - title: String type: string - additionalProperties: true title: Object type: object - items: {} title: Array type: array description: >- What true means. A string, an object or an array. type: object instructions: anyOf: - title: String type: string - additionalProperties: true title: Object type: object - items: {} title: Array type: array description: >- The evaluation prompt for this question. A string, an object or an array. Native OpenAI also accepts arrays of input_text, input_image and input_json parts. Question parts cannot contain prompt cache breakpoints. type: enum: - noul type: string required: - type - instructions title: NoulQuestion type: object - description: >- Picks one of the given options and returns the probability of each. properties: criteria: additionalProperties: anyOf: - title: String type: string - additionalProperties: true title: Object type: object - items: {} title: Array type: array - title: 'Null' type: 'null' description: The option description, or null. description: >- Options keyed by name, each mapped to a description or null. A description can be a string, an object or an array. minProperties: 1 type: object instructions: anyOf: - title: String type: string - additionalProperties: true title: Object type: object - items: {} title: Array type: array description: >- The evaluation prompt for this question. A string, an object or an array. Native OpenAI also accepts arrays of input_text, input_image and input_json parts. Question parts cannot contain prompt cache breakpoints. type: enum: - choice type: string required: - type - instructions - criteria title: ChoiceQuestion type: object - description: >- Places the state on an ordered scale and returns the probability of each level. properties: criteria: description: >- Ordered level descriptions from lowest to highest. At least two levels. Each level can be a string, an object or an array. items: {} minItems: 2 type: array instructions: anyOf: - title: String type: string - additionalProperties: true title: Object type: object - items: {} title: Array type: array description: >- The evaluation prompt for this question. A string, an object or an array. Native OpenAI also accepts arrays of input_text, input_image and input_json parts. Question parts cannot contain prompt cache breakpoints. type: enum: - score type: string required: - type - instructions - criteria title: ScoreQuestion type: object - description: >- Beta, TypeSafe jev-preview only. Locates one target in an image from assets and returns its bounding box. Always returns a box; pair it with a noul question to check presence. properties: instructions: anyOf: - title: String type: string - additionalProperties: true title: Object type: object - items: {} title: Array type: array description: >- The evaluation prompt for this question. A string, an object or an array. Native OpenAI also accepts arrays of input_text, input_image and input_json parts. Question parts cannot contain prompt cache breakpoints. type: enum: - box type: string required: - type - instructions title: BoxQuestion type: object description: >- Typed questions keyed by an identifier of your choice. Each answer is returned under the same key. Native OpenAI accepts 1 to 400 questions. minProperties: 1 type: object retry: $ref: '#/components/schemas/ClassifyRetryConfig' description: Retry configuration for the request. state: anyOf: - title: String type: string - additionalProperties: true title: Object type: object - items: {} title: Array type: array description: >- The content to evaluate. A string, an object or an array. For TypeSafe models, reference images declared in assets by writing their reference string. Native OpenAI accepts messages and function/custom-tool history. User messages support input_text, input_image and input_json; images can use inline base64 data URLs or public HTTP(S) URLs. At most 128 images across state and question instructions. Shared text/image parts support explicit prompt_cache_breakpoint markers. File IDs, audio, item references and bare content parts are unsupported. Other objects and arrays are serialized as text. timeout: $ref: '#/components/schemas/ClassifyTimeoutConfig' description: >- Timeout for each model call, including every retry and fallback. A timed-out call returns 408 and advances to the next fallback after configured retries are exhausted. required: - model - state - questions type: object required: true responses: '200': content: application/json: examples: openai_answers: summary: OpenAI Decisions answers and usage value: model: openai/gpt-6-luna answers: positive: type: noul noul: 1 rating: type: score score: 2 legend: '0': Negative '1': Neutral '2': Positive probabilities: '0': 0 '1': 0 '2': 1 confidence: 1 sentiment: type: choice choice: positive probabilities: negative: 0 neutral: 0 positive: 1 confidence: 1 usage: input_tokens_details: cached_tokens: 0 cache_write_tokens: 0 input_tokens: 408 output_tokens: 0 input_cost: 0.0000408 output_cost: 0 total_cost: 0.0000408 partial_refusal: description: >- Illustrative response. A provider may refuse one question while answering the rest. Refusals have no scored fields. summary: One question is refused value: model: openai/gpt-6-luna answers: positive: type: refusal rating: type: score score: 2 legend: '0': Negative '1': Neutral '2': Positive probabilities: '0': 0 '1': 0 '2': 1 confidence: 1 sentiment: type: choice choice: positive probabilities: negative: 0 neutral: 0 positive: 1 confidence: 1 usage: input_tokens_details: cached_tokens: 0 cache_write_tokens: 0 input_tokens: 408 output_tokens: 0 input_cost: 0.0000408 output_cost: 0 total_cost: 0.0000408 schema: additionalProperties: false properties: answers: additionalProperties: $ref: '#/components/schemas/ClassifyAnswer' description: >- Answers keyed by the question identifiers from the request. type: object model: description: >- The requested ID of the model that answered. This can be a fallback model. type: string telemetry: $ref: '#/components/schemas/ResponseTelemetry' description: OpenTelemetry trace and span identifiers for this request. usage: $ref: '#/components/schemas/ClassifyUsage' description: The usage information for the request. required: - model - answers - usage type: object description: Returns one answer per question. '400': content: application/json: schema: type: object additionalProperties: false properties: error: $ref: '#/components/schemas/APIError' required: - error description: >- Malformed JSON, missing or unsupported model, or invalid retry, timeout, or fallback fields. '401': content: application/json: schema: type: object additionalProperties: false properties: error: $ref: '#/components/schemas/APIError' required: - error description: Missing, invalid, expired or revoked API key. '403': content: application/json: schema: type: object additionalProperties: false properties: error: $ref: '#/components/schemas/APIError' required: - error description: >- The API key lacks classify permission, or the workspace or project cannot access the model. '408': content: application/json: schema: type: object additionalProperties: false properties: error: $ref: '#/components/schemas/APIError' required: - error description: >- The final model call exceeded its timeout after available retries and fallbacks were exhausted. '422': content: application/json: schema: type: object additionalProperties: false properties: error: $ref: '#/components/schemas/APIError' required: - error description: The state or a question violates the classification contract. '429': content: application/json: schema: type: object additionalProperties: false properties: error: $ref: '#/components/schemas/APIError' required: - error description: A plan rate limit, budget or provider rate limit was exceeded. '500': content: application/json: schema: type: object additionalProperties: false properties: error: $ref: '#/components/schemas/APIError' required: - error description: An internal model-resolution error occurred. '502': content: application/json: schema: type: object additionalProperties: false properties: error: $ref: '#/components/schemas/APIError' required: - error description: >- The upstream provider failed or returned an invalid classification response. security: - ApiKey: [] x-code-samples: - label: Python lang: python source: |- result = client.router.decisions.create( model="typesafe/jev-latest", state="The parcel arrived two days late and the box was crushed.", questions={ "is_complaint": {"type": "noul", "instructions": "Is the customer complaining?"}, "topic": { "type": "choice", "instructions": "What is the message mainly about?", "criteria": {"delivery": "Shipping or delivery issues", "product": "Product quality", "other": None}, }, "severity": {"type": "score", "instructions": "How severe is the issue?", "criteria": ["Minor", "Moderate", "Severe"]}, }, ) - label: 'cURL: OpenAI text' lang: bash source: |- curl https://router.toqan.ai/eu/v3/router/decisions \ -H "Authorization: Bearer ${TOQAN_API_KEY}" \ -H "Content-Type: application/json" \ --data-binary @- <<'JSON' { "model": "openai/gpt-6-luna", "questions": { "positive": { "criteria": { "false": "The customer is unhappy.", "true": "The customer is happy." }, "instructions": "Is the sentiment positive?", "type": "noul" }, "rating": { "criteria": [ "Negative", "Neutral", "Positive" ], "instructions": "Rate sentiment.", "type": "score" }, "sentiment": { "criteria": { "negative": "Negative sentiment", "neutral": null, "positive": "Positive sentiment" }, "instructions": "Classify sentiment.", "type": "choice" } }, "state": "The customer says: I love this product. It is wonderful!" } JSON - label: 'cURL: OpenAI inline image' lang: bash source: |- curl https://router.toqan.ai/eu/v3/router/decisions \ -H "Authorization: Bearer ${TOQAN_API_KEY}" \ -H "Content-Type: application/json" \ --data-binary @- <<'JSON' { "model": "openai/gpt-6-luna", "questions": { "contains_text": { "instructions": "Does the image contain text?", "type": "noul" } }, "state": [ { "content": [ { "text": "Evaluate this image.", "type": "input_text" }, { "image_url": "data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAoAAAAKCAYAAACNMs+9AAAACXBIWXMAAAsTAAALEwEAmpwYAAAAAXNSR0IArs4c6QAAAARnQU1BAACxjwv8YQUAAACLSURBVHgBdY8NDYAgEIXBBEQgAhG0gRGMYBNtoA2M4EygDYygDfCxvdMbk7d9A453f8ZQMcYAWuBVzIHujeGyxE8nYz7dJWYl01p749zxDKABEzhYfDNaMPZSAUymJLZLuvSsSVUhx4G7VM2xpSzQ5YYhLUHDCGoad47ixXjxY84qi1a9QPgZpdYLPVkbtsfywz3jAAAAAElFTkSuQmCC", "type": "input_image" } ], "role": "user" } ] } JSON - label: 'cURL: Fallbacks, retries and identity' lang: bash source: |- curl https://router.toqan.ai/eu/v3/router/decisions \ -H "Authorization: Bearer ${TOQAN_API_KEY}" \ -H "Content-Type: application/json" \ --data-binary @- <<'JSON' { "fallbacks": [ { "model": "openai/gpt-5.6-luna" } ], "identity": { "id": "customer-demo", "display_name": "Sample customer" }, "model": "openai/gpt-6-luna", "questions": { "positive": { "criteria": { "false": "The customer is unhappy.", "true": "The customer is happy." }, "instructions": "Is the sentiment positive?", "type": "noul" }, "rating": { "criteria": [ "Negative", "Neutral", "Positive" ], "instructions": "Rate sentiment.", "type": "score" }, "sentiment": { "criteria": { "negative": "Negative sentiment", "neutral": null, "positive": "Positive sentiment" }, "instructions": "Classify sentiment.", "type": "choice" } }, "retry": { "count": 2, "on_codes": [ 429, 502, 503, 504 ] }, "state": "The customer says: I love this product. It is wonderful!", "timeout": { "call_timeout": 2000 } } JSON components: schemas: ClassifyAsset: additionalProperties: false properties: data_base64: description: >- Standard base64 with padding, without whitespace or a data: prefix. PNG or JPEG still images, 256 to 1568 px per side, at most 16 MiB each. type: string format: description: The image format. enum: - png - jpg - jpeg type: string type: description: The asset type. Only image is supported. enum: - image type: string required: - data_base64 - format type: object ClassifyCacheConfig: additionalProperties: false properties: ttl: description: >- Time to live for the cached response in seconds, up to 259200 (3 days). Defaults to 3600. format: int64 maximum: 259200 minimum: 1 type: integer type: description: Cache type. Only exact_match is supported. enum: - exact_match type: string required: - type type: object FallbackConfig: additionalProperties: false properties: model: type: string required: - model type: object ResponseIdentity: additionalProperties: false properties: display_name: type: string email: type: string id: type: string metadata: items: type: object additionalProperties: {} type: - array - 'null' tags: items: type: string type: - array - 'null' required: - id type: object ClassifyPromptCacheOptions: additionalProperties: false properties: mode: description: >- Implicit uses the shared input boundary by default; explicit uses only marked input boundaries. enum: - implicit - explicit type: string ttl: description: >- Minimum cache lifetime. Only 30m is supported by native OpenAI Decisions. enum: - 30m type: string type: object ClassifyRetryConfig: additionalProperties: false properties: count: description: >- Number of retries per model after the initial attempt (0-5). Set 0 to disable retries. When retry is omitted, the gateway retries twice. format: int64 maximum: 5 minimum: 0 type: integer on_codes: description: >- HTTP status codes that trigger retries, between 100 and 599. Defaults to [408, 429, 500, 502, 503, 504, 529]. Retry-After is honored; otherwise the gateway backs off exponentially. items: format: int64 maximum: 599 minimum: 100 type: integer type: array required: - count type: object ClassifyTimeoutConfig: additionalProperties: false properties: call_timeout: description: >- Timeout in milliseconds for each model call. Set 2000 for two seconds. When retry is configured, timeout errors (408) are automatically retried. format: int64 minimum: 1 type: integer required: - call_timeout type: object ClassifyAnswer: description: >- One answer to a classification question. The type selects the answer fields. A refusal contains only type. discriminator: mapping: box: '#/components/schemas/ClassifyBoxAnswer' choice: '#/components/schemas/ClassifyChoiceAnswer' noul: '#/components/schemas/ClassifyNoulAnswer' refusal: '#/components/schemas/ClassifyRefusalAnswer' score: '#/components/schemas/ClassifyScoreAnswer' propertyName: type oneOf: - $ref: '#/components/schemas/ClassifyNoulAnswer' - $ref: '#/components/schemas/ClassifyChoiceAnswer' - $ref: '#/components/schemas/ClassifyScoreAnswer' - $ref: '#/components/schemas/ClassifyBoxAnswer' - $ref: '#/components/schemas/ClassifyRefusalAnswer' ResponseTelemetry: additionalProperties: false properties: span_id: type: string trace_id: type: string required: - trace_id - span_id type: object ClassifyUsage: additionalProperties: false properties: input_cost: description: >- Cost (USD) of input tokens. Present when billing was computed for this request. format: double type: number input_tokens: description: The number of input tokens processed. format: int64 type: integer input_tokens_details: $ref: '#/components/schemas/ClassifyInputTokensDetails' description: >- Provider-reported cache token usage. OpenAI Decisions charges cache reads and writes at the model cache rates; these tokens are included in input_tokens. output_cost: description: >- Cost (USD) of output tokens. 0 for native classify providers, including OpenAI Decisions. Present when billing was computed for this request. format: double type: number output_tokens: description: >- The number of output tokens generated. Free for native classify providers, billed at the model rate for emulated chat models. format: int64 type: integer total_cost: description: >- Total cost (USD) of the request. Present when billing was computed for this request. format: double type: number required: - input_tokens - output_tokens type: object APIError: additionalProperties: false properties: code: type: - string - 'null' failures: {} message: type: string param: type: - string - 'null' type: type: string required: - message - type - param - code type: object ClassifyBoxAnswer: additionalProperties: false properties: box: $ref: '#/components/schemas/ClassifyBox' description: The bounding box of the target. Present for box answers. confidence: description: >- How sure the model is of the answer it selected: on emulated chat models the largest value in probabilities, so it adds nothing the distribution does not; native providers return their own confidence value, which can sit below the largest probability. Present for choice, score and box answers. format: double type: number type: enum: - box type: string required: - type - box title: ClassifyBoxAnswer type: object ClassifyChoiceAnswer: additionalProperties: false properties: choice: description: The selected option. Present for choice answers. type: string confidence: description: >- How sure the model is of the answer it selected: on emulated chat models the largest value in probabilities, so it adds nothing the distribution does not; native providers return their own confidence value, which can sit below the largest probability. Present for choice, score and box answers. format: double type: number probabilities: additionalProperties: type: number format: double description: >- Probability distribution over the options or levels. Present for choice and score answers. type: object type: enum: - choice type: string required: - type - choice title: ClassifyChoiceAnswer type: object ClassifyNoulAnswer: additionalProperties: false properties: noul: description: >- Probability between 0 and 1 that the statement holds. Present for noul answers. format: double type: number type: enum: - noul type: string required: - type - noul title: ClassifyNoulAnswer type: object ClassifyRefusalAnswer: additionalProperties: false properties: type: enum: - refusal type: string required: - type title: ClassifyRefusalAnswer type: object ClassifyScoreAnswer: additionalProperties: false properties: confidence: description: >- How sure the model is of the answer it selected: on emulated chat models the largest value in probabilities, so it adds nothing the distribution does not; native providers return their own confidence value, which can sit below the largest probability. Present for choice, score and box answers. format: double type: number legend: additionalProperties: type: string description: Level index to level description. Present for score answers. type: object probabilities: additionalProperties: type: number format: double description: >- Probability distribution over the options or levels. Present for choice and score answers. type: object score: description: >- Position on the scale, not an index: on emulated chat models and OpenAI Decisions the weighted index, each level index multiplied by that level probability and summed, so the value is usually fractional; typesafe/jev-latest returns its own score. Present for score answers. format: double type: number type: enum: - score type: string required: - type - score title: ClassifyScoreAnswer type: object ClassifyInputTokensDetails: additionalProperties: false properties: cache_write_tokens: description: Input tokens written to cache. format: int64 type: integer cached_tokens: description: Input tokens read from cache. format: int64 type: integer required: - cached_tokens - cache_write_tokens type: object ClassifyBox: additionalProperties: false properties: bottom: description: >- Distance of the bottom edge from the top of the image, as a fraction of the image height. format: double maximum: 1 minimum: 0 type: number left: description: >- Distance of the left edge from the left of the image, as a fraction of the image width. format: double maximum: 1 minimum: 0 type: number right: description: >- Distance of the right edge from the left of the image, as a fraction of the image width. format: double maximum: 1 minimum: 0 type: number top: description: >- Distance of the top edge from the top of the image, as a fraction of the image height. format: double maximum: 1 minimum: 0 type: number required: - top - bottom - left - right type: object securitySchemes: ApiKey: type: http scheme: bearer bearerFormat: JWT ```` This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.