fix backend to support gpt-5-mini

app/main.py

@@ -29,11 +29,38 @@ except ImportError:
 API_KEY = os.getenv("GEMINI_API_KEY", "")
 MODEL   = os.getenv("GEMINI_MODEL", "gemini-2.5-pro")
 GEMINI_SMALL_MODEL = os.getenv("GEMINI_SMALL_MODEL")
-OPENAI_SMALL_MODEL = os.getenv("OPENAI_SMALL_MODEL") or "gpt-4o-mini"
+DEFAULT_OPENAI_SMALL_MODEL = "gpt-4o-mini"
+OPENAI_SMALL_MODEL = os.getenv("OPENAI_SMALL_MODEL") or DEFAULT_OPENAI_SMALL_MODEL
 OPENAI_API_KEY = os.getenv("OPENAI_API_KEY")
 USE_OPENAI = os.getenv("USE_OPENAI", "").lower() == "true"
 PORT    = int(os.getenv("PORT", "7860"))
 
+_OPENAI_RESPONSES_MODELS_ENV = os.getenv("OPENAI_RESPONSES_MODELS", "")
+RESPONSES_API_MODEL_NAMES = {"gpt-5-mini"}
+if _OPENAI_RESPONSES_MODELS_ENV:
+    RESPONSES_API_MODEL_NAMES.update(
+        model.strip().lower()
+        for model in _OPENAI_RESPONSES_MODELS_ENV.split(",")
+        if model.strip()
+    )
+
+_OPENAI_RESPONSES_PREFIXES_ENV = os.getenv("OPENAI_RESPONSES_PREFIXES", "")
+_RESPONSES_API_MODEL_PREFIXES = ["gpt-5"]
+if _OPENAI_RESPONSES_PREFIXES_ENV:
+    _RESPONSES_API_MODEL_PREFIXES.extend(
+        prefix.strip().lower()
+        for prefix in _OPENAI_RESPONSES_PREFIXES_ENV.split(",")
+        if prefix.strip()
+    )
+RESPONSES_API_MODEL_PREFIXES = tuple(_RESPONSES_API_MODEL_PREFIXES)
+RESPONSES_API_ERROR_HINTS = (
+    "only supported in v1/responses",
+    "use the responses api",
+    "use the responses endpoint",
+    "please call the responses api",
+    "please use the responses endpoint",
+)
+
 gemini_client = genai.Client(api_key=API_KEY) if API_KEY else None
 gpt_client = OpenAI(api_key=OPENAI_API_KEY) if (OPENAI_API_KEY and OpenAI and USE_OPENAI) else None
 
@@ -162,6 +189,22 @@ def _build_responses_input(system: str, contents: Any) -> List[Dict[str, Any]]:
         {"role": "user", "content": _build_openai_content(contents, for_chat=False)},
     ]
 
+def _requires_responses_api(model: str) -> bool:
+    lowered = (model or "").lower()
+    if not lowered:
+        return False
+    if lowered in RESPONSES_API_MODEL_NAMES:
+        return True
+    return any(
+        prefix and lowered.startswith(prefix)
+        for prefix in RESPONSES_API_MODEL_PREFIXES
+    )
+
+
+def _should_use_responses_fallback(err: Exception) -> bool:
+    message = str(err).lower()
+    return any(hint in message for hint in RESPONSES_API_ERROR_HINTS)
+
 
 def _extract_chat_content(resp: Any) -> str:
     content = resp.choices[0].message.content
@@ -199,17 +242,20 @@ def _invoke_gpt_model(model: str, system: str, contents: Any) -> str:
     if not gpt_client:
         raise RuntimeError("GPT client is not configured")
     messages = _build_chat_messages(system, contents)
+    responses_input: Optional[List[Dict[str, Any]]] = None
+    if _requires_responses_api(model):
+        responses_input = _build_responses_input(system, contents)
+        resp = gpt_client.responses.create(model=model, input=responses_input)
+        return _extract_responses_content(resp)
     try:
         resp = gpt_client.chat.completions.create(model=model, messages=messages)
         return _extract_chat_content(resp)
     except Exception as err:
-        message = str(err)
-        if "only supported in v1/responses" not in message:
+        if not _should_use_responses_fallback(err):
             raise
-        resp = gpt_client.responses.create(
-            model=model,
-            input=_build_responses_input(system, contents),
-        )
+        if responses_input is None:
+            responses_input = _build_responses_input(system, contents)
+        resp = gpt_client.responses.create(model=model, input=responses_input)
         return _extract_responses_content(resp)