diff --git a/agent/auxiliary_client.py b/agent/auxiliary_client.py
index 05dee5b342b4..1b3658677c0f 100644
--- a/agent/auxiliary_client.py
+++ b/agent/auxiliary_client.py
@@ -1066,7 +1066,10 @@ class _CodexCompletionsAdapter:
 
             _host_src = str(getattr(self._client, "base_url", "") or "")
             _is_xai = base_url_host_matches(_host_src, "x.ai") or base_url_host_matches(_host_src, "api.x.ai")
-            _is_github = base_url_host_matches(_host_src, "githubcopilot.com")
+            _is_github = (
+                base_url_host_matches(_host_src, "githubcopilot.com")
+                or base_url_host_matches(_host_src, "models.github.ai")
+            )
             _is_codex_backend = (
                 base_url_host_matches(_host_src, "chatgpt.com")
                 and "/backend-api/codex" in _host_src.lower()
diff --git a/tests/agent/test_auxiliary_client.py b/tests/agent/test_auxiliary_client.py
index b047b0d2a5eb..72d39031ad33 100644
--- a/tests/agent/test_auxiliary_client.py
+++ b/tests/agent/test_auxiliary_client.py
@@ -4989,6 +4989,17 @@ class TestCodexAdapterPromptCacheKey:
         ])
         assert "prompt_cache_retention" not in captured
 
+    def test_prompt_cache_retention_skipped_for_github_models_host(self):
+        """models.github.ai is a GitHub Responses host in the main transport
+        (agent/chat_completion_helpers.py) — the auxiliary path must exclude
+        it from cache-retention emission the same way as githubcopilot.com."""
+        adapter, captured = self._build_adapter(base_url="https://models.github.ai/inference")
+        adapter.create(messages=[
+            {"role": "system", "content": "SYS"},
+            {"role": "user", "content": "hi"},
+        ])
+        assert "prompt_cache_retention" not in captured
+
 
 class TestCodexAdapterGithubResponsesMessageIdDrop:
     """_CodexCompletionsAdapter must drop codex_message_items ``id`` when