Update app.js

app.js

@@ -68,12 +68,15 @@ app.post('/api/generate', async (req, res) => {
         } else {
             // Handles Claude Sonnet, Claude Haiku, and Llama Maverick
             const bedrockModelId = getBedrockModelId(model);
-            const command = new ConverseCommand({
+       
+          /*   const command = new ConverseCommand({
                 modelId: bedrockModelId,
                 system: [{ text: system_prompt || CLAUDE_SYSTEM_PROMPT }], 
                 messages: [{ role: "user", content: [{ text: prompt }] }],
                
-              inferenceConfig:  model.includes("claude") ? {  maxTokens: 48000, temperature: 1 }: {maxTokens: 3800, temperature: 1 } ,
+              inferenceConfig:  model.includes("claude") ? {  maxTokens: 48000, temperature: 1 }:
+                (model.includes("haiku")? {maxTokens: 30000, temperature: 1 } 
+                 :  {maxTokens: 3800, temperature: 1 }),
            
               performanceConfig: model.includes("maverick") ? {latency: "standard"} : undefined,
                 
@@ -82,7 +85,40 @@ app.post('/api/generate', async (req, res) => {
                     output_config: { effort: "high" }
                 } : undefined // Llama does not support Claude's specific thinking fields
             });
-            
+
+            */
+
+const command = new ConverseCommand({
+    modelId: bedrockModelId,
+    system: [{ text: system_prompt || CLAUDE_SYSTEM_PROMPT }],
+    messages: [{ role: "user", content: [{ text: prompt }] }],
+    
+    // Ensure maxTokens is large enough for reasoning + response
+    inferenceConfig: { 
+        maxTokens: model.includes("haiku") ? 32000 : 4000, 
+        temperature: 1 
+    },
+
+    performanceConfig: model.includes("maverick") ? { latency: "standard" } : undefined,
+
+    additionalModelRequestFields: (function() {
+        if (model.includes("haiku")) {
+            return {
+                reasoning_config: {
+                    type: "enabled",
+                    budget_tokens: 2048 // As seen in your screenshot
+                }
+            };
+        } else if (model.includes("claude")) {
+            return {
+                thinking: { type: "adaptive" },
+                output_config: { effort: "high" }
+            };
+        }
+        return undefined;
+    })()
+});
+          
             const response = await bedrockClient.send(command);
             const text = response.output.message.content.find(b => b.text)?.text;
             const tokenUsage = response.usage ? (response.usage.inputTokens + response.usage.outputTokens) : 0;