diff --git a/.speakeasy/in.openapi.yaml b/.speakeasy/in.openapi.yaml index 14af9a4..1560110 100755 --- a/.speakeasy/in.openapi.yaml +++ b/.speakeasy/in.openapi.yaml @@ -5051,6 +5051,8 @@ components: - $ref: '#/components/schemas/ParetoRouterPlugin' - $ref: '#/components/schemas/FusionPlugin' type: array + prediction: + $ref: '#/components/schemas/Prediction' presence_penalty: description: Presence penalty (-2.0 to 2.0) example: 0 @@ -17926,6 +17928,7 @@ components: - logit_bias - logprobs - top_logprobs + - prediction - seed - response_format - structured_outputs @@ -18234,6 +18237,45 @@ components: - context_compression example: guardrail type: string + Prediction: + description: >- + Static predicted output content. Supported models can use this to reduce latency when much of the response is + known in advance. + example: + content: Expected response + type: content + nullable: true + properties: + content: + anyOf: + - type: string + - items: + $ref: '#/components/schemas/PredictionContentText' + type: array + type: + enum: + - content + type: string + required: + - type + - content + type: object + PredictionContentText: + description: Text content part for a predicted output. + example: + text: Expected response + type: text + properties: + text: + type: string + type: + enum: + - text + type: string + required: + - type + - text + type: object PreferredMaxLatency: anyOf: - format: double