[aisdk] add Truncation: {auto, default} as an encoding metadata option

aisdk/ai/provider/openai/internal/codec/encode.go

@@ -118,10 +118,8 @@ func applyCallOptions(params *responses.ResponseNewParams, opts api.CallOptions,
 	// Apply provider options from metadata
 	applyProviderMetadata(params, opts)
 
-	// Apply model-specific settings
-	if modelConfig.RequiredAutoTruncation {
-		params.Truncation = "auto"
-	}
+	// Apply truncation settings: caller-provided takes precedence over model defaults
+	applyTruncationSettings(params, opts, modelConfig)
 
 	// Apply reasoning settings and handle unsupported options for reasoning models
 	reasoningWarnings := applyReasoningSettings(params, opts, modelConfig)
@@ -185,6 +183,24 @@ func applyProviderMetadata(params *responses.ResponseNewParams, opts api.CallOpt
 	}
 }
 
+// applyTruncationSettings applies truncation settings to the parameters.
+// Caller-provided truncation via metadata takes precedence over model defaults.
+func applyTruncationSettings(params *responses.ResponseNewParams, opts api.CallOptions, modelConfig modelConfig) {
+	// Check if caller provided a truncation setting via metadata
+	if opts.ProviderMetadata != nil {
+		metadata := GetMetadata(&opts)
+		if metadata != nil && metadata.Truncation != "" {
+			params.Truncation = responses.ResponseNewParamsTruncation(metadata.Truncation)
+			return
+		}
+	}
+
+	// Fall back to model-specific defaults
+	if modelConfig.RequiredAutoTruncation {
+		params.Truncation = "auto"
+	}
+}
+
 // applyReasoningSettings applies settings specific to reasoning models
 // and handles unsupported options
 func applyReasoningSettings(params *responses.ResponseNewParams, opts api.CallOptions,

aisdk/ai/provider/openai/internal/codec/metadata.go

@@ -66,6 +66,16 @@ type Metadata struct {
 	// Supported values are `concise` and `detailed`.
 	ReasoningSummary string `json:"reasoning_summary,omitempty"`
 
+	// Truncation specifies the truncation strategy to use for the model input.
+	// When set, this overrides any model-specific default truncation behavior.
+	//
+	// Supported values are `auto` and `disabled` (default).
+	// - `auto`: If the context of this response and previous ones exceeds the
+	//   model's context window size, the model will truncate the response to fit.
+	// - `disabled`: If the context exceeds the model's context window size,
+	//   the request will fail with a 400 error.
+	Truncation string `json:"truncation,omitempty"`
+
 	// --- Used in blocks ---
 
 	// ImageDetail indicates the level of detail that should be used when processing