diff --git a/studio/backend/routes/inference.py b/studio/backend/routes/inference.py
index fdf9fd8f5e..6d05be2310 100644
--- a/studio/backend/routes/inference.py
+++ b/studio/backend/routes/inference.py
@@ -1959,8 +1959,11 @@ async def openai_chat_completions(
     Supports multimodal messages: ``content`` may be a plain string or a
     list of content parts (``text`` / ``image_url``).
 
-    Streaming (default):  returns SSE chunks matching OpenAI's format.
-    Non-streaming:        returns a single ChatCompletion JSON object.
+    Non-streaming (default): returns a single ChatCompletion JSON object.
+    Streaming:               returns SSE chunks matching OpenAI's format.
+
+    ``stream`` defaults to ``false`` to match OpenAI's spec; clients opt
+    into SSE by sending ``stream: true``.
 
     Automatically routes to the correct backend:
     - GGUF models → llama-server via LlamaCppBackend
@@ -2184,6 +2187,9 @@ async def openai_chat_completions(
 
         cancel_event = threading.Event()
         completion_id = f"chatcmpl-{uuid.uuid4().hex[:12]}"
+        # `stream` defaults to False on ChatCompletionRequest (OpenAI spec
+        # parity). Naive curl / .NET / System.Text.Json clients omitting
+        # the field used to get SSE here and choke on deserialization (#5047).
         if payload.stream:
             return await _openai_passthrough_stream(
                 request,
diff --git a/studio/backend/tests/test_openai_tool_passthrough.py b/studio/backend/tests/test_openai_tool_passthrough.py
index a379282b70..e87faa3b32 100644
--- a/studio/backend/tests/test_openai_tool_passthrough.py
+++ b/studio/backend/tests/test_openai_tool_passthrough.py
@@ -299,11 +299,57 @@ class TestChatCompletionRequestToolFields:
     def test_stream_defaults_false_matching_openai_spec(self):
         # OpenAI's /v1/chat/completions spec defaults `stream` to false.
         # Studio previously defaulted to true, which broke naive curl
-        # clients that omit `stream` (they expect a JSON blob, got SSE).
+        # clients (and .NET / System.Text.Json SDKs per #5047) that omit
+        # `stream` -- they expect a JSON blob, got SSE.
         # Pin the corrected default so it can't silently regress.
         req = self._make()
         assert req.stream is False
 
+    def test_post_without_stream_field_decodes_to_stream_false_over_http(
+        self, monkeypatch
+    ):
+        # Wire-level guard for the same default: a POST body that omits
+        # `stream` entirely (the exact shape naive curl / .NET clients
+        # send) must deserialise into stream=False *and* the response
+        # must be `application/json`, never `text/event-stream`.
+        # Mounts the real `routes.inference.router` so this catches
+        # regressions in middleware/aliasing on the actual endpoint
+        # (e.g. someone adding a request layer that injects stream=True
+        # before pydantic builds the model). Backends are bypassed by
+        # routing through `provider_type` and stubbing the external
+        # provider proxy.
+        from fastapi import FastAPI
+        from fastapi.responses import JSONResponse
+        from fastapi.testclient import TestClient
+
+        import routes.inference as inference_route
+        from auth.authentication import get_current_subject
+
+        captured = {}
+
+        async def _fake_proxy(payload, request):
+            captured["stream"] = payload.stream
+            return JSONResponse({"choices": [], "object": "chat.completion"})
+
+        monkeypatch.setattr(inference_route, "_proxy_to_external_provider", _fake_proxy)
+
+        app = FastAPI()
+        app.include_router(inference_route.router)
+        app.dependency_overrides[get_current_subject] = lambda: "test-user"
+
+        client = TestClient(app)
+        resp = client.post(
+            "/chat/completions",
+            json = {
+                "messages": [{"role": "user", "content": "hi"}],
+                "provider_type": "openai",
+            },
+        )
+        assert resp.status_code == 200
+        assert resp.headers["content-type"].startswith("application/json")
+        assert "text/event-stream" not in resp.headers["content-type"]
+        assert captured["stream"] is False
+
     def test_multiturn_tool_loop_messages(self):
         req = ChatCompletionRequest(
             messages = [