diff --git a/backend/src/proxy/openai/routes.rs b/backend/src/proxy/openai/routes.rs
index 42409a032..8c36db1bb 100644
--- a/backend/src/proxy/openai/routes.rs
+++ b/backend/src/proxy/openai/routes.rs
@@ -1,6 +1,7 @@
 //! OpenAI 兼容 API 路由。
 
 use axum::{
+    extract::DefaultBodyLimit,
     routing::{get, post},
     Router,
 };
@@ -22,4 +23,5 @@ pub fn router() -> Router<AppState> {
         .route("/v1/models/catalog", get(model_catalog))
         .route("/v1/models/{model_id}", get(model_detail))
         .route("/v1/models/{model_id}/info", get(model_info))
+        .layer(DefaultBodyLimit::disable())
 }
diff --git a/backend/tests/proxy/openai/responses_routes.rs b/backend/tests/proxy/openai/responses_routes.rs
index ae7efcb85..6ce88a396 100644
--- a/backend/tests/proxy/openai/responses_routes.rs
+++ b/backend/tests/proxy/openai/responses_routes.rs
@@ -59,6 +59,30 @@ async fn responses_route_should_reject_unknown_models_with_openai_error() {
     );
 }
 
+#[tokio::test]
+async fn responses_route_should_accept_body_larger_than_axum_default_limit() {
+    let (app, api_key, _dir) = test_app_with_client_api_key().await;
+    let large_input = "x".repeat(3 * 1024 * 1024);
+    let body = serde_json::json!({
+        "model": "unknown-model",
+        "input": large_input,
+    })
+    .to_string();
+    let response = app
+        .oneshot(
+            Request::builder()
+                .method("POST")
+                .uri("/v1/responses")
+                .header("authorization", format!("Bearer {api_key}"))
+                .header("content-type", "application/json")
+                .body(Body::from(body))
+                .unwrap(),
+        )
+        .await
+        .unwrap();
+    assert_eq!(response.status(), StatusCode::NOT_FOUND);
+}
+
 #[tokio::test]
 async fn responses_route_should_terminate_generated_stream_errors_with_done_marker() {
     let (app, api_key, _dir) = test_app_with_client_api_key().await;