diff --git a/src/config.js b/src/config.js
index b59fead..1ba4f8e 100644
--- a/src/config.js
+++ b/src/config.js
@@ -68,7 +68,14 @@ export async function getSystemMessage() {
 export const VLLM_ENDPOINTS = [
 	{ id: "hermes.ai.unturf.com", url: "https://hermes.ai.unturf.com/v1" },
 	{ id: "hermes2.ai.unturf.com", url: "https://hermes2.ai.unturf.com/v1" },
-	{ id: "qwen.ai.unturf.com", url: "https://qwen.ai.unturf.com/v1" },
+	{
+		id: "qwen.ai.unturf.com",
+		url: "https://qwen.ai.unturf.com/v1",
+		// Ollama endpoints don't return max_model_len, so we specify it here
+		modelContextWindows: {
+			'hf.co/unsloth/Qwen3-Coder-30B-A3B-Instruct-GGUF:Q4_K_M': 32768,
+		}
+	},
 ];
 
 // Function to get API configuration (custom or default)
diff --git a/src/models.js b/src/models.js
index 30f6dd9..94635cc 100644
--- a/src/models.js
+++ b/src/models.js
@@ -87,7 +87,13 @@ export async function fetchModelsFromEndpoints() {
 			return models.map((model) => {
 				// Get context window size from API or use known values for Groq models
 				let maxTokens = model.max_tokens || model.context_length || model.max_context_length || model.max_model_len;
-				
+
+				// Check if endpoint has modelContextWindows config (for Ollama/other endpoints without API max_model_len)
+				if (endpoint.modelContextWindows && endpoint.modelContextWindows[model.id]) {
+					maxTokens = endpoint.modelContextWindows[model.id];
+					console.log(`Using configured context window for model ${model.id}: ${maxTokens} tokens`);
+				}
+
 				// Hardcoded context windows for known Groq models (since API doesn't return them)
 				if (endpoint.id.includes('groq.com')) {
 					const groqContextWindows = {