From afebcb34556db2250f0ab3829d7d8665e947683d Mon Sep 17 00:00:00 2001
From: Nate Armstrong <narmstrong@berri.ai>
Date: Mon, 19 Oct 2026 12:55:27 -0700
Subject: [PATCH] fix(gateway): translated chat answers carry a chatcmpl id and
 object (#47366)

The provider configs leave id out because Python mints it on its own
ModelResponse, so non-streaming Anthropic, Bedrock, Vertex and Gemini
answers reached gateway clients with no id and no object. The chat
handler now names each translated answer as Python does: a minted
chatcmpl- id, or Gemini's responseId, and object chat.completion. It does
so before the answer is cached, so a cache hit replays the same id.
---
 .../crates/gateway/tests/chat_translation.rs  | 17 ++++++++++++++
 .../crates/inference-chat/src/handler.rs      | 15 ++++++++++++
 .../vertex_and_google_ai_studio_gemini.rs     | 23 +++++++++++++++++--
 3 files changed, 53 insertions(+), 2 deletions(-)

diff --git a/litellm-rust/crates/gateway/tests/chat_translation.rs b/litellm-rust/crates/gateway/tests/chat_translation.rs
index 3c7f7acf305..cbf3a0ce0b1 100644
--- a/litellm-rust/crates/gateway/tests/chat_translation.rs
+++ b/litellm-rust/crates/gateway/tests/chat_translation.rs
@@ -474,6 +474,23 @@ async fn bedrock_signs_with_the_keys_region_and_endpoint_in_the_config() {
     );
 }
 
+#[rstest]
+#[case::anthropic("claude")]
+#[case::bedrock("bedrock")]
+#[tokio::test]
+async fn a_translated_answer_carries_a_chat_completion_id_and_object(#[case] model: &str) {
+    let harness = harness().await;
+    let answer = harness
+        .json(json!({"model": model, "messages": answered("tooluse_1"), "max_tokens": 16}))
+        .await;
+
+    assert!(
+        answer["id"].as_str().unwrap().starts_with("chatcmpl-"),
+        "{answer}"
+    );
+    assert_eq!(answer["object"], "chat.completion");
+}
+
 #[tokio::test]
 async fn bedrock_tool_round_trip() {
     let harness = harness().await;
diff --git a/litellm-rust/crates/inference-chat/src/handler.rs b/litellm-rust/crates/inference-chat/src/handler.rs
index 95764ccda42..160cc9f9221 100644
--- a/litellm-rust/crates/inference-chat/src/handler.rs
+++ b/litellm-rust/crates/inference-chat/src/handler.rs
@@ -123,6 +123,7 @@ pub(super) async fn execute(
             })?;
             config
                 .transform_response(&model, ProviderChatResponseData { body }, &stream_shape)
+                .map(named)
                 .map_err(Error::from)
                 .map_err(as_response_error)
         },
@@ -130,6 +131,20 @@ pub(super) async fn execute(
     .await
 }
 
+/// The answer named as Python's `ModelResponse` names it: by the id the config kept (Gemini's
+/// `responseId`), else a minted `chatcmpl-` id, with `object: chat.completion`. Named before it
+/// is cached, so a cache hit replays the id the answer was first sent with.
+fn named(mut response: ChatCompletionsResponse) -> ChatCompletionsResponse {
+    response
+        .extra
+        .entry("id")
+        .or_insert_with(|| format!("chatcmpl-{}", uuid::Uuid::new_v4()).into());
+    response
+        .extra
+        .insert("object".into(), "chat.completion".into());
+    response
+}
+
 /// Re-tag an error raised while normalizing a response the provider already
 /// returned.
 ///
diff --git a/litellm-rust/crates/llms/src/vertex_ai/gemini/vertex_and_google_ai_studio_gemini.rs b/litellm-rust/crates/llms/src/vertex_ai/gemini/vertex_and_google_ai_studio_gemini.rs
index 891fff963bf..40d0c102934 100644
--- a/litellm-rust/crates/llms/src/vertex_ai/gemini/vertex_and_google_ai_studio_gemini.rs
+++ b/litellm-rust/crates/llms/src/vertex_ai/gemini/vertex_and_google_ai_studio_gemini.rs
@@ -1461,7 +1461,14 @@ pub fn transform_generate_content_response(
     }
     let usage = calculate_usage(&body)?;
     let [grounding, url_context, safety, citations] = metadata;
-    let extra = Map::from_iter([
+    // Python names the answer by Gemini's `responseId` when there is one; the host mints an id
+    // otherwise.
+    let response_id = body
+        .get("responseId")
+        .and_then(Value::as_str)
+        .filter(|id| !id.is_empty())
+        .map(|id| ("id".to_owned(), Value::from(id)));
+    let extra = Map::from_iter(response_id.into_iter().chain([
         (
             "vertex_ai_grounding_metadata".to_owned(),
             Value::Array(grounding),
@@ -1475,7 +1482,7 @@ pub fn transform_generate_content_response(
             "vertex_ai_citation_metadata".to_owned(),
             Value::Array(citations),
         ),
-    ]);
+    ]));
     Ok(ChatCompletionsResponse {
         created: unix_now(),
         model,
@@ -2060,6 +2067,18 @@ mod tests {
             .unwrap()
     }
 
+    #[rstest]
+    #[case::named(json!("resp_1"), Some(json!("resp_1")))]
+    #[case::unnamed(Value::Null, None)]
+    fn the_answer_is_named_by_the_response_id(#[case] id: Value, #[case] expected: Option<Value>) {
+        let answer = response(json!({
+            "responseId": id,
+            "candidates": [{"content": {"role": "model", "parts": [{"text": "Hi"}]}, "finishReason": "STOP"}],
+            "usageMetadata": {"promptTokenCount": 1, "candidatesTokenCount": 1, "totalTokenCount": 2}
+        }));
+        assert_eq!(answer.get("id").cloned(), expected);
+    }
+
     #[rstest]
     fn text_reasoning_and_usage_come_back_in_openai_shape() {
         let answer = response(json!({