diff --git a/langsmith-fleet-worker-attempt.json b/langsmith-fleet-worker-attempt.json
index aca5d45e3..901cef897 100644
--- a/langsmith-fleet-worker-attempt.json
+++ b/langsmith-fleet-worker-attempt.json
@@ -1,13 +1,13 @@
 {
   "agent": "codex",
   "cli_version": "0.125.0",
-  "emitted_at": "2026-07-24T21:53:26.737479Z",
+  "emitted_at": "2026-07-24T22:55:53.506408Z",
   "execution_profile": "codex-default",
   "fallback_models": [
     "gpt-5.4"
   ],
   "operation_role": "worker",
-  "pr_number": "2822",
+  "pr_number": "2825",
   "requested_model": "gpt-5.5",
   "runner": "reusable-codex-run",
   "schema": "langsmith-fleet/v1",
diff --git a/tests/scripts/test_task_decomposer.py b/tests/scripts/test_task_decomposer.py
index 6c92b13b7..2be4ba8bc 100644
--- a/tests/scripts/test_task_decomposer.py
+++ b/tests/scripts/test_task_decomposer.py
@@ -508,10 +508,13 @@ def test_get_llm_client_github_token_defaults(monkeypatch) -> None:
     client, provider = client_info
     assert provider == "github-models"
     assert isinstance(client, FakeChatOpenAI)
+    from tools.langchain_client import _github_model_id
     from tools.llm_provider import GITHUB_MODELS_BASE_URL
     from tools.llm_registry import configured_model_for_provider
 
-    assert client.kwargs["model"] == configured_model_for_provider("github-models")
+    assert client.kwargs["model"] == _github_model_id(
+        configured_model_for_provider("github-models")
+    )
     assert client.kwargs["base_url"] == GITHUB_MODELS_BASE_URL
 
 
diff --git a/tests/tools/test_langchain_client.py b/tests/tools/test_langchain_client.py
index ccdfd4a76..006b245ed 100644
--- a/tests/tools/test_langchain_client.py
+++ b/tests/tools/test_langchain_client.py
@@ -158,7 +158,8 @@ def test_explicit_github_provider_uses_reviewed_model_when_model_is_omitted(
     assert resolved.provider == langchain_client.PROVIDER_GITHUB
     assert resolved.model == "github-reviewed"
     assert isinstance(resolved.client, FakeChatOpenAI)
-    assert resolved.client.kwargs["model"] == "github-reviewed"
+    # github-models GA namespaces bare ids with the openai/ publisher
+    assert resolved.client.kwargs["model"] == "openai/github-reviewed"
 
 
 def test_build_chat_client_env_model_override(monkeypatch: pytest.MonkeyPatch) -> None:
@@ -175,7 +176,7 @@ def test_build_chat_client_env_model_override(monkeypatch: pytest.MonkeyPatch) -
     assert resolved is not None
     assert resolved.model == "gpt-4o-mini"
     assert isinstance(resolved.client, FakeChatOpenAI)
-    assert resolved.client.kwargs["model"] == "gpt-4o-mini"
+    assert resolved.client.kwargs["model"] == "openai/gpt-4o-mini"
 
 
 def test_build_chat_client_blocked_model_override_does_not_shift_provider(
@@ -801,11 +802,43 @@ def test_build_openai_client_normal_model_has_temperature(
     assert resolved.client.kwargs["temperature"] == 0.1
 
 
+class _CaptureChatOpenAI:
+    def __init__(self, **kwargs):
+        self.kwargs = kwargs
+
+
 class _CaptureChatAnthropic:
     def __init__(self, **kwargs):
         self.kwargs = kwargs
 
 
+def test_github_model_id_namespaces_bare_ids():
+    assert langchain_client._github_model_id("codex-mini-latest") == "openai/codex-mini-latest"
+    assert langchain_client._github_model_id("gpt-5") == "openai/gpt-5"
+
+
+def test_github_model_id_leaves_namespaced_ids_unchanged():
+    assert langchain_client._github_model_id("openai/gpt-5") == "openai/gpt-5"
+    assert (
+        langchain_client._github_model_id("mistral-ai/mistral-large") == "mistral-ai/mistral-large"
+    )
+
+
+def test_build_github_client_sends_namespaced_model():
+    client = langchain_client._build_github_client(
+        _CaptureChatOpenAI, model="codex-mini-latest", token="t", timeout=30, max_retries=2
+    )
+    assert client.kwargs["model"] == "openai/codex-mini-latest"
+    assert client.kwargs["base_url"] == langchain_client.GITHUB_MODELS_BASE_URL
+
+
+def test_build_github_client_preserves_already_namespaced():
+    client = langchain_client._build_github_client(
+        _CaptureChatOpenAI, model="openai/gpt-5", token="t", timeout=30, max_retries=2
+    )
+    assert client.kwargs["model"] == "openai/gpt-5"
+
+
 def _build_anthropic(model: str):
     return langchain_client._build_anthropic_client(
         _CaptureChatAnthropic, model=model, token="t", timeout=30, max_retries=2
diff --git a/tools/langchain_client.py b/tools/langchain_client.py
index cf1d97ce4..48a6e6039 100644
--- a/tools/langchain_client.py
+++ b/tools/langchain_client.py
@@ -200,11 +200,26 @@ def _build_anthropic_client(
     return chat_anthropic(**kwargs)
 
 
+def _github_model_id(model: str) -> str:
+    """Return a GitHub Models GA-compatible model id (``publisher/model``).
+
+    The GA endpoint (``models.github.ai/inference``) requires publisher-namespaced ids
+    such as ``openai/gpt-5``. A bare id (e.g. ``codex-mini-latest``) 404s ("page not
+    found") there. Default a bare id to the ``openai/`` publisher; already-namespaced
+    ids (containing ``/``) are returned unchanged. (If a bare id maps to a model the
+    GitHub Models catalog does not offer at all, the durable fix is an owner-reviewed
+    change of the github-models selection to a catalogued id — this only normalizes
+    the id format.)
+    """
+    name = model.strip()
+    return name if "/" in name else f"openai/{name}"
+
+
 def _build_github_client(
     chat_openai: type, *, model: str, token: str, timeout: int, max_retries: int
 ) -> object:
     kwargs: dict = {
-        "model": model,
+        "model": _github_model_id(model),
         "base_url": GITHUB_MODELS_BASE_URL,
         "api_key": token,
         "timeout": timeout,