diff --git a/server-rs/crates/api-server/src/editor_project_icon_specs.rs b/server-rs/crates/api-server/src/editor_project_icon_specs.rs
index fdd85d2a8..eafff07d1 100644
--- a/server-rs/crates/api-server/src/editor_project_icon_specs.rs
+++ b/server-rs/crates/api-server/src/editor_project_icon_specs.rs
@@ -5,7 +5,7 @@ use axum::{
     extract::{Extension, State},
     http::StatusCode,
 };
-use platform_llm::{LlmMessage, LlmRunRequest};
+use platform_llm::{EDITOR_AGENT_GPT5_MODEL, LlmMessage, LlmRunRequest};
 use serde::{Deserialize, Deserializer, Serialize, Serializer, de};
 use serde_json::{Value, json};
 use shared_contracts::assets::EditorCanvasGenerationCompletionPayload;
@@ -260,15 +260,17 @@ async fn run_icon_spec_extra_param_llm(
     state: &AppState,
     prompt: String,
 ) -> Result<ExtraParam, AppError> {
-    let configured_llm_client = state.llm_client().ok_or_else(|| {
+    let configured_llm_client = state.vector_engine_llm_client().ok_or_else(|| {
         AppError::from_status(StatusCode::SERVICE_UNAVAILABLE).with_details(json!({
             "provider": "editor-icon-spec-llm",
-            "message": "服务端尚未配置可用的 LLM API Key",
+            "message": "服务端尚未配置可用的 VectorEngine LLM API Key",
         }))
     })?;
     let retry_backoff_ms = configured_llm_client.config().retry_backoff_ms();
     let llm_client = configured_llm_client.clone().with_max_retries(0);
-    let request = LlmRunRequest::new(vec![LlmMessage::user(prompt)]);
+    let request = LlmRunRequest::new(vec![LlmMessage::user(prompt)])
+        .with_model(EDITOR_AGENT_GPT5_MODEL)
+        .with_openai_chat();
 
     for attempt in 1..=ICON_SPEC_LLM_MAX_ATTEMPTS {
         match llm_client.run(request.clone()).await {
diff --git a/server-rs/crates/api-server/src/llm/icon_specs.rs b/server-rs/crates/api-server/src/llm/icon_specs.rs
index f8591159d..9ab92221d 100644
--- a/server-rs/crates/api-server/src/llm/icon_specs.rs
+++ b/server-rs/crates/api-server/src/llm/icon_specs.rs
@@ -3,7 +3,7 @@ use axum::{
     extract::{Extension, State},
     http::StatusCode,
 };
-use platform_llm::{LlmMessage, LlmRunRequest};
+use platform_llm::{EDITOR_AGENT_GPT5_MODEL, LlmMessage, LlmRunRequest};
 use serde::Deserialize;
 use serde_json::{Value, json};
 use std::time::Duration;
@@ -144,15 +144,17 @@ fn normalize_refined_text(value: &str) -> Result<String, RefinedTextError> {
 }
 
 async fn run_refine_text_llm(state: &AppState, prompt: String) -> Result<String, AppError> {
-    let configured_llm_client = state.llm_client().ok_or_else(|| {
+    let configured_llm_client = state.vector_engine_llm_client().ok_or_else(|| {
         AppError::from_status(StatusCode::SERVICE_UNAVAILABLE).with_details(json!({
             "provider": "editor-icon-spec-llm",
-            "message": "服务端尚未配置可用的 LLM API Key",
+            "message": "服务端尚未配置可用的 VectorEngine LLM API Key",
         }))
     })?;
     let retry_backoff_ms = configured_llm_client.config().retry_backoff_ms();
     let llm_client = configured_llm_client.clone().with_max_retries(0);
-    let request = LlmRunRequest::new(vec![LlmMessage::user(prompt)]);
+    let request = LlmRunRequest::new(vec![LlmMessage::user(prompt)])
+        .with_model(EDITOR_AGENT_GPT5_MODEL)
+        .with_openai_chat();
 
     for attempt in 1..=ICON_SPEC_LLM_MAX_ATTEMPTS {
         match llm_client.run(request.clone()).await {