Merge branch 'fix-issue-5344-quantization-guardrail' into feat-gemma4-moe-4bit-swap

studio/backend/core/inference/external_provider.py

@@ -29,6 +29,7 @@ logger = structlog.get_logger(__name__)
 # still accept it. Match the 4-7 line specifically so we keep the knob
 # live on every other Claude generation.
 _ANTHROPIC_TOP_K_DEPRECATED = re.compile(r"^claude-(?:opus|sonnet|haiku)-4-7(?:[-.]|$)")
+_OPENAI_REASONING_SUMMARY_UNSUPPORTED = re.compile(r"^o3(?:[-.]|$)")
 
 
 class _AnthropicThinkingSpec(NamedTuple):
@@ -1659,6 +1660,9 @@ class ExternalProviderClient:
         # to wrap, and the chat reasoning panel stays blank. Always pair
         # an explicit effort with summary except for the explicit "off"
         # case (effort: "none"), where summaries are pointless.
+        summary_unsupported = bool(
+            _OPENAI_REASONING_SUMMARY_UNSUPPORTED.match(model.strip().lower())
+        )
         if reasoning_effort in (
             "minimal",
             "low",
@@ -1667,11 +1671,15 @@ class ExternalProviderClient:
             "max",
             "xhigh",
         ):
-            body["reasoning"] = {"effort": reasoning_effort, "summary": "auto"}
+            body["reasoning"] = {"effort": reasoning_effort}
+            if not summary_unsupported:
+                body["reasoning"]["summary"] = "auto"
         elif reasoning_effort == "none" or enable_thinking is False:
             body["reasoning"] = {"effort": "none"}
         elif enable_thinking is True:
-            body["reasoning"] = {"effort": "medium", "summary": "auto"}
+            body["reasoning"] = {"effort": "medium"}
+            if not summary_unsupported:
+                body["reasoning"]["summary"] = "auto"
         if instructions_parts:
             body["instructions"] = "\n\n".join(instructions_parts)
         if max_tokens is not None:

studio/backend/tests/test_openai_responses_translation.py

@@ -286,6 +286,68 @@ def test_responses_reasoning_effort_included_when_requested(monkeypatch):
     assert captured["body"]["reasoning"] == {"effort": "high", "summary": "auto"}
 
 
+def test_responses_reasoning_summary_omitted_for_o3(monkeypatch):
+    captured: dict = {}
+
+    def handler(request: httpx.Request) -> httpx.Response:
+        captured["body"] = json.loads(request.content.decode("utf-8"))
+        return httpx.Response(
+            200,
+            content = _responses_sse([{"type": "response.completed", "response": {}}]),
+            headers = {"content-type": "text/event-stream"},
+        )
+
+    _mock_http_client(monkeypatch, handler)
+
+    async def run():
+        client = _make_client()
+        async for _ in client._stream_openai_responses(
+            messages = [{"role": "user", "content": "hi"}],
+            model = "o3",
+            temperature = 0.7,
+            top_p = 0.95,
+            max_tokens = None,
+            enable_thinking = None,
+            reasoning_effort = "high",
+        ):
+            pass
+        await client.close()
+
+    _drive(run())
+    assert captured["body"]["reasoning"] == {"effort": "high"}
+
+
+def test_responses_reasoning_summary_omitted_for_o3_with_enable_thinking(monkeypatch):
+    captured: dict = {}
+
+    def handler(request: httpx.Request) -> httpx.Response:
+        captured["body"] = json.loads(request.content.decode("utf-8"))
+        return httpx.Response(
+            200,
+            content = _responses_sse([{"type": "response.completed", "response": {}}]),
+            headers = {"content-type": "text/event-stream"},
+        )
+
+    _mock_http_client(monkeypatch, handler)
+
+    async def run():
+        client = _make_client()
+        async for _ in client._stream_openai_responses(
+            messages = [{"role": "user", "content": "hi"}],
+            model = "o3",
+            temperature = 0.7,
+            top_p = 0.95,
+            max_tokens = None,
+            enable_thinking = True,
+            reasoning_effort = None,
+        ):
+            pass
+        await client.close()
+
+    _drive(run())
+    assert captured["body"]["reasoning"] == {"effort": "medium"}
+
+
 def test_responses_reasoning_effort_none_omits_summary(monkeypatch):
     captured: dict = {}