Switch from Claude Haiku to Claude Sonnet for better accuracy

Haiku was being too literal in price extraction, often "correcting"
valid alternative prices (like Amazon other sellers) to the main
buy box price.

Sonnet is more capable and should handle nuanced cases better.

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

Author: clucraft

backend/src/services/ai-extractor.ts

@@ -173,7 +173,7 @@ async function extractWithAnthropic(
   const preparedHtml = prepareHtmlForAI(html);
 
   const response = await anthropic.messages.create({
-    model: 'claude-3-haiku-20240307',
+    model: 'claude-sonnet-4-20250514',
     max_tokens: 1024,
     messages: [
       {
@@ -269,7 +269,7 @@ async function verifyWithAnthropic(
     .replace('$CURRENCY$', currency) + preparedHtml;
 
   const response = await anthropic.messages.create({
-    model: 'claude-3-haiku-20240307',
+    model: 'claude-sonnet-4-20250514',
     max_tokens: 512,
     messages: [{ role: 'user', content: prompt }],
   });
@@ -625,7 +625,7 @@ async function arbitrateWithAnthropic(
   const prompt = ARBITRATION_PROMPT.replace('$CANDIDATES$', candidatesList) + preparedHtml;
 
   const response = await anthropic.messages.create({
-    model: 'claude-3-haiku-20240307',
+    model: 'claude-sonnet-4-20250514',
     max_tokens: 512,
     messages: [{ role: 'user', content: prompt }],
   });