diff --git a/backend/cpp/llama-cpp/Makefile b/backend/cpp/llama-cpp/Makefile
index 5c02e597b..0fdac506b 100644
--- a/backend/cpp/llama-cpp/Makefile
+++ b/backend/cpp/llama-cpp/Makefile
@@ -1,5 +1,5 @@
 
-LLAMA_VERSION?=f2cc7282ce0042641df55e0e563b75e824e2b647
+LLAMA_VERSION?=a0e6dc7384edb8f5d927155e1d26c10b4a310e6d
 LLAMA_REPO?=https://github.com/ggerganov/llama.cpp
 
 CMAKE_ARGS?=
diff --git a/backend/cpp/llama-cpp/grpc-server.cpp b/backend/cpp/llama-cpp/grpc-server.cpp
index f6bc0f34d..cc27a40f6 100644
--- a/backend/cpp/llama-cpp/grpc-server.cpp
+++ b/backend/cpp/llama-cpp/grpc-server.cpp
@@ -108,6 +108,12 @@ using grpc::Status;
 // requests without a matching "authorization: Bearer <token>" metadata header.
 
 // Cached auth token — empty means auth is disabled.
+// Newer llama.cpp returns common_chat_params::prompt as a common_chat_prompt
+// (use .str()); older forks such as turboquant still return std::string.
+static inline std::string chat_prompt_str(const std::string & p) { return p; }
+template <typename T>
+static inline std::string chat_prompt_str(const T & p) { return p.str(); }
+
 static std::string g_grpc_auth_token;
 
 // Minimal constant-time comparison (avoids OpenSSL dependency)
@@ -3823,7 +3829,7 @@ public:
             dummy_inputs.use_jinja = params_base.use_jinja;
             
             const auto rendered = common_chat_templates_apply(ctx_server.impl->chat_params.tmpls.get(), dummy_inputs);
-            rendered_template = rendered.prompt;
+            rendered_template = chat_prompt_str(rendered.prompt);
         }
         
         response->set_rendered_template(rendered_template);