feat: switch to MiniCPM5-1B-Claude-Opus-Fable5-V2-Thinking-Q8_0 GGUF

- Updated model path + model ID for new 1B thinking model
- Updated prompt builder to use MiniCPM5 native /think trigger
- Updated clean_text to strip MiniCPM5 special tokens
- Bumped n_ctx from 2048 to 8192

src/application/chat/use_cases.rs

@@ -89,8 +89,8 @@ pub fn build_prompt(messages: &[ChatMessage], tools: &Option<Vec<ToolDef>>) -> S
         }
     }
 
-    // Generation prompt: non-thinking mode
-    prompt.push_str("<|im_start|>assistant\n<think>\n\n</think>\n\n");
+    // Generation prompt: thinking mode (MiniCPM5 native)
+    prompt.push_str("<|im_start|>assistant\n/think\n\n");
     prompt
 }
 
@@ -182,8 +182,13 @@ pub fn build_sampler(params: &SamplerParams) -> LlamaSampler {
 pub fn clean_text(text: &str) -> String {
     text.replace("<|im_end|>", "")
         .replace("<|im_start|>", "")
-        .replace("<think>", "")
-        .replace("</think>", "")
+        .replace("<|thought_begin|>", "")
+        .replace("<|thought_end|>", "")
+        .replace("<|tool_call|>", "")
+        .replace("<|execute_start|>", "")
+        .replace("<|execute_end|>", "")
+        .replace("/think", "")
+        .replace("/no_think", "")
         .trim()
         .to_string()
 }

src/config/mod.rs

@@ -4,8 +4,8 @@
 
 use std::sync::LazyLock;
 
-const DEFAULT_MODEL_PATH: &str = "/models/MiniCPM-V-4.6-Q4_K_M.gguf";
-pub const MODEL_ID: &str = "minicpm-v-4.6";
+const DEFAULT_MODEL_PATH: &str = "/models/MiniCPM5-1B-Claude-Opus-Fable5-V2-Thinking-Q8_0.gguf";
+pub const MODEL_ID: &str = "minicpm5-1b-fable5-v2-thinking";
 
 /// Application configuration loaded at startup from environment variables.
 #[derive(Debug, Clone)]
@@ -44,7 +44,7 @@ impl AppConfig {
                 .and_then(|v| v.parse().ok())
                 .unwrap_or(8080),
             log_level: std::env::var("RUST_LOG").unwrap_or_else(|_| "info".to_string()),
-            n_ctx: 2048,
+            n_ctx: 8192,
             n_batch: 512,
             n_threads: 4,
         }

src/presentation/handler/health.rs

@@ -8,6 +8,6 @@ use crate::domain::entity::HealthResponse;
 pub async fn health_check() -> Json<HealthResponse> {
     Json(HealthResponse {
         status: "ok".into(),
-        model: format!("{MODEL_ID}-q4_k_m"),
+        model: format!("{MODEL_ID}-q8_0"),
     })
 }