fix: remove topP parameter from Bedrock inference config (#8388)

Co-authored-by: Matt Rubens <[email protected]>

Author: daniel-lxs

src/api/providers/__tests__/bedrock-reasoning.spec.ts

@@ -182,7 +182,7 @@ describe("AwsBedrockHandler - Extended Thinking", () => {
 			)
 		})
 
-		it("should include topP when thinking is disabled", async () => {
+		it("should not include topP when thinking is disabled (global removal)", async () => {
 			handler = new AwsBedrockHandler({
 				apiProvider: "bedrock",
 				apiModelId: "anthropic.claude-3-7-sonnet-20250219-v1:0",
@@ -216,10 +216,10 @@ describe("AwsBedrockHandler - Extended Thinking", () => {
 				chunks.push(chunk)
 			}
 
-			// Verify that topP IS present when thinking is disabled
+			// Verify that topP is NOT present for any model (removed globally)
 			expect(mockSend).toHaveBeenCalledTimes(1)
 			expect(capturedPayload).toBeDefined()
-			expect(capturedPayload.inferenceConfig).toHaveProperty("topP", 0.1)
+			expect(capturedPayload.inferenceConfig).not.toHaveProperty("topP")
 
 			// Verify that additionalModelRequestFields is not present or empty
 			expect(capturedPayload.additionalModelRequestFields).toBeUndefined()

src/api/providers/bedrock.ts

@@ -45,7 +45,6 @@ import type { SingleCompletionHandler, ApiHandlerCreateMessageMetadata } from ".
 interface BedrockInferenceConfig {
 	maxTokens: number
 	temperature?: number
-	topP?: number
 }
 
 // Define interface for Bedrock additional model request fields
@@ -374,12 +373,8 @@ export class AwsBedrockHandler extends BaseProvider implements SingleCompletionH
 			maxTokens: modelConfig.maxTokens || (modelConfig.info.maxTokens as number),
 			temperature: modelConfig.temperature ?? (this.options.modelTemperature as number),
 		}
-
-		if (!thinkingEnabled) {
-			inferenceConfig.topP = 0.1
-		}
-
-		// Check if 1M context is enabled for Claude Sonnet 4 / 4.5
+	
+		// Check if 1M context is enabled for Claude Sonnet 4
 		// Use parseBaseModelId to handle cross-region inference prefixes
 		const baseModelId = this.parseBaseModelId(modelConfig.id)
 		const is1MContextEnabled =
@@ -649,7 +644,6 @@ export class AwsBedrockHandler extends BaseProvider implements SingleCompletionH
 			const inferenceConfig: BedrockInferenceConfig = {
 				maxTokens: modelConfig.maxTokens || (modelConfig.info.maxTokens as number),
 				temperature: modelConfig.temperature ?? (this.options.modelTemperature as number),
-				...(thinkingEnabled ? {} : { topP: 0.1 }), // Only set topP when thinking is NOT enabled
 			}
 
 			// For completePrompt, use a unique conversation ID based on the prompt