diff --git a/src/translate-modal.js b/src/translate-modal.js
index 9134f49..7975f9a 100644
--- a/src/translate-modal.js
+++ b/src/translate-modal.js
@@ -336,8 +336,12 @@ export function openTranslateModal() {
 			// Calculate actual tokens that will be sent to AI (including preservation and chunking)
 			let totalTokens = 0;
 			
+			// Import all needed functions
+			const { getSelectedModelMaxTokens, getSelectedModel } = await import("./models.js");
+			const { countTokensWithUpstream } = await import("./token_estimator.js");
+			const { preserveSpecialContent, estimateProcessingTime, SUPPORTED_LANGUAGES } = await import("./translation.js");
+			
 			// Simulate the same chunking logic that translateHTML will use
-			const { getSelectedModelMaxTokens } = await import("./models.js");
 			const maxTokens = getSelectedModelMaxTokens() || 8192;
 			const reserveTokens = 1000;
 			const availableTokens = Math.max(maxTokens - reserveTokens, 2000);
diff --git a/src/translation.js b/src/translation.js
index 608243c..4fad771 100644
--- a/src/translation.js
+++ b/src/translation.js
@@ -125,12 +125,12 @@ async function* sendMessageWithHistory(messageHistory) {
 	const modelMaxTokens = getSelectedModelMaxTokens();
 	const inputText = messageHistory.map(msg => msg.content).join('');
 	const inputTokens = Math.ceil(inputText.length / 4); // Simple token estimation
-	const buffer = Math.max(2048, Math.floor(inputTokens * 1.5)); // Use 1.5x input tokens as buffer, minimum 2048
-	const availableTokens = Math.max(100, modelMaxTokens - inputTokens - buffer);
+	const reserveTokens = 1000; // Reserve for output tokens
+	const availableTokens = Math.max(100, modelMaxTokens - inputTokens - reserveTokens);
 	
 	console.log("Translation using endpoint:", apiUrl);
 	console.log("Translation using model:", model);
-	console.log("Translation token calculation: max=" + modelMaxTokens + ", input≈" + inputTokens + ", buffer=" + buffer + ", available≈" + availableTokens);
+	console.log("Translation token calculation: max=" + modelMaxTokens + ", input≈" + inputTokens + ", reserve=" + reserveTokens + ", available≈" + availableTokens);
 
 	const response = await fetch(apiUrl, {
 		method: "POST",
@@ -542,8 +542,8 @@ export async function translateHTML(htmlContent, targetLanguage, sourceUrl = nul
 		console.log(`Content requires chunking: ${totalEstimatedTokens} total tokens > ${availableTokens} limit`);
 		
 		// Split the original HTML content into semantic chunks
-		// Reduce chunk size to account for translation expansion
-		const maxChunkInputTokens = Math.floor(availableTokens / 2.8); // Account for 1.8x expansion + overhead
+		// Use reasonable chunk size for the model's context window
+		const maxChunkInputTokens = Math.floor((maxTokens - 1500) / 2.5); // Leave room for prompt + expansion
 		const chunks = await splitHTMLIntoChunks(processedContent, maxChunkInputTokens, getSelectedModel());
 		console.log(`Split into ${chunks.length} chunks for parallel translation`);