Merge pull request #532 from MaEscalanteHe/feature/update-openai-models

fix: update OpenAI models and send supported reasoning parameters

AppImage/config/verified_ai_models.json

@@ -38,31 +38,24 @@
   },
   "openai": {
     "models": [
-      "gpt-4.1-nano-2025-04-14",
+      "gpt-6-luna",
-      "gpt-4.1-nano",
+      "gpt-5.6-luna",
-      "gpt-4.1-2025-04-14",
+      "gpt-5.4-mini",
-      "gpt-4.1",
+      "gpt-5.4-mini-2026-03-17",
-      "gpt-4o-2024-11-20",
+      "gpt-6-sol",
-      "gpt-4o-2024-05-13",
+      "gpt-6.1-sol",
-      "gpt-4o-mini",
+      "gpt-5.6-terra",
-      "gpt-3.5-turbo-16k",
+      "gpt-5.2",
-      "gpt-4o-mini-2024-07-18",
+      "gpt-5.2-2025-12-11",
-      "gpt-5-nano",
+      "gpt-5.4",
-      "gpt-3.5-turbo-1106",
+      "gpt-5.4-2026-03-05",
-      "gpt-3.5-turbo-0125",
+      "gpt-5.6-sol",
-      "gpt-5-nano-2025-08-07",
+      "gpt-5.5",
-      "gpt-4.1-mini",
+      "gpt-5.5-2026-04-23",
-      "gpt-4.1-mini-2025-04-14",
+      "gpt-6-astra"
-      "gpt-3.5-turbo",
-      "gpt-4o",
-      "gpt-4o-2024-08-06",
-      "gpt-4",
-      "gpt-4-turbo",
-      "gpt-4-turbo-2024-04-09",
-      "gpt-4-0613"
     ],
-    "recommended": "gpt-4.1-nano",
+    "recommended": "gpt-6-luna",
-    "_note": "Verified 2026-09-02 with the OpenAI API — 22 of 74 tested pass. Includes stable aliases (gpt-4.1-nano, gpt-4o-mini, ...), dated snapshots for pinning (gpt-4o-2024-11-20, gpt-4.1-nano-2025-04-14) and legacy families (gpt-3.5-turbo, gpt-4) for cost-optimised use cases. Recommended kept as `gpt-4.1-nano` (stable alias, 1.39s)."
+    "_note": "Verified 2026-10-26 against the live API. Recommended `gpt-6-luna` (cheapest current model). Dropped every model that is retired (gpt-4.1-nano, gpt-3.5-turbo, gpt-4, gpt-4-turbo, ...) or has a shutdown date (gpt-4.1, gpt-4o, gpt-4o-mini, gpt-5-nano, gpt-5.1, gpt-5.4-nano)."
   },
   "anthropic": {
     "models": [

AppImage/scripts/ai_providers/openai_provider.py

@@ -5,6 +5,7 @@ Models are loaded dynamically from the API.
 """
 from typing import Optional, List
 import json
+import re
 import urllib.request
 import urllib.error
 from .base import AIProvider, AIProviderError
@@ -36,11 +37,11 @@ class OpenAIProvider(AIProvider):
     ]
     
     # Recommended models for chat (in priority order)
-    RECOMMENDED_PREFIXES = ['gpt-4o-mini', 'gpt-4o', 'gpt-4-turbo', 'gpt-4', 'gpt-3.5-turbo']
+    RECOMMENDED_PREFIXES = ['gpt-6-luna', 'gpt-5.6-luna', 'gpt-5.4-mini', 'gpt-6-sol']
 
     @staticmethod
     def _is_reasoning_model(model: str) -> bool:
-        """True for OpenAI reasoning models (o-series + non-chat gpt-5+).
+        """True for OpenAI reasoning models (o-series + non-chat gpt-5 and later).
 
         These use a stricter API contract than chat models:
           - Must use ``max_completion_tokens`` instead of ``max_tokens``
@@ -54,11 +55,24 @@ class OpenAIProvider(AIProvider):
         # o1, o3, o4, o5 ...  (o<digit>...)
         if len(m) >= 2 and m[0] == 'o' and m[1].isdigit():
             return True
-        # gpt-5, gpt-5-mini, gpt-5.1, gpt-5.2-pro ...  EXCEPT *-chat-latest
+        # gpt-5, gpt-5-mini, gpt-5.1, gpt-6-luna, gpt-6.1-sol ...  EXCEPT *-chat-latest
-        if m.startswith('gpt-5') and '-chat' not in m:
+        major = re.match(r'gpt-(\d+)', m)
-            return True
+        return bool(major) and int(major.group(1)) >= 5 and '-chat' not in m
-        return False
+
-    
+    @staticmethod
+    def _reasoning_effort(model: str) -> str:
+        """Lowest reasoning_effort the model accepts.
+
+        gpt-5 / gpt-5-mini / gpt-5-nano (and the o-series) accept 'minimal'.
+        gpt-5.1 and later reject 'minimal' with HTTP 400, and gpt-6-astra /
+        gpt-6.1-sol also reject 'none', so 'low' is the value all of them
+        accept. In tests it spent 0-5 reasoning tokens on a notification.
+        """
+        m = model.lower()
+        if re.match(r'gpt-5(?!\.)', m) or m.startswith('o'):
+            return 'minimal'
+        return 'low'
+
     def list_models(self) -> List[str]:
         """List available models for chat completions.
 
@@ -189,22 +203,19 @@ class OpenAIProvider(AIProvider):
         # apply model-specific defaults.
         if self.base_url:
             payload['max_tokens'] = max_tokens
-        # Reasoning models (o1/o3/o4/gpt-5*, excluding *-chat-latest) use a
+        # Reasoning models (o1/o3/o4/gpt-5+, excluding *-chat-latest) use a
         # different parameter contract: max_completion_tokens instead of
         # max_tokens, and no temperature field. Sending the classic chat
         # parameters to them produces HTTP 400 Bad Request.
         #
         # They also spend output budget on internal reasoning by default,
         # which empties the user-visible reply when max_tokens is small
-        # (like the ~200 we use for notifications). reasoning_effort
+        # (like the ~200 we use for notifications). The lowest
-        # 'minimal' keeps that internal reasoning to a minimum so the
+        # reasoning_effort the model accepts keeps that internal reasoning
-        # entire budget is available for the translation, which is
+        # to a minimum so the budget goes to the translation.
-        # exactly what this pipeline wants. OpenAI documents 'minimal',
-        # 'low', 'medium', 'high' — 'minimal' is the right setting for a
-        # straightforward translate+explain task.
         elif self._is_reasoning_model(self.model):
             payload['max_completion_tokens'] = max_tokens
-            payload['reasoning_effort'] = 'minimal'
+            payload['reasoning_effort'] = self._reasoning_effort(self.model)
         else:
             payload['max_tokens'] = max_tokens
             payload['temperature'] = 0.3

AppImage/scripts/tests/test_openai_compatible_ai_models.py

@@ -98,6 +98,31 @@ class OpenAICompatibleModelTests(unittest.TestCase):
         self.assertNotIn("temperature", provider.captured_payload)
         self.assertNotIn("max_tokens", provider.captured_payload)
 
+    def test_newer_reasoning_models_use_low_effort(self):
+        # gpt-5.1+ reject 'minimal'; gpt-6* were sent chat parameters before.
+        for model in ("gpt-5.6-luna", "gpt-6-luna", "gpt-6.1-sol"):
+            with self.subTest(model=model):
+                provider = CapturingOpenAIProvider(api_key="token", model=model)
+
+                self.assertEqual(provider.generate("system", "user", max_tokens=123), "ok")
+
+                self.assertEqual(provider.captured_payload["max_completion_tokens"], 123)
+                self.assertEqual(provider.captured_payload["reasoning_effort"], "low")
+                self.assertNotIn("temperature", provider.captured_payload)
+                self.assertNotIn("max_tokens", provider.captured_payload)
+
+    def test_classic_chat_models_keep_chat_contract(self):
+        for model in ("gpt-4.1", "gpt-4o-mini"):
+            with self.subTest(model=model):
+                provider = CapturingOpenAIProvider(api_key="token", model=model)
+
+                self.assertEqual(provider.generate("system", "user", max_tokens=123), "ok")
+
+                self.assertEqual(provider.captured_payload["max_tokens"], 123)
+                self.assertEqual(provider.captured_payload["temperature"], 0.3)
+                self.assertNotIn("reasoning_effort", provider.captured_payload)
+                self.assertNotIn("max_completion_tokens", provider.captured_payload)
+
     def test_runtime_ai_config_prefers_provider_specific_model(self):
         manager = self._manager({
             "ai_enabled": "true",

web/app/[locale]/docs/monitor/notifications/page.tsx

@@ -814,7 +814,7 @@ curl -H "Authorization: Bearer <api-token>" \\
 curl -X POST http://<host>:8008/api/notifications/test-ai \\
   -H "Authorization: Bearer <api-token>" \\
   -H "Content-Type: application/json" \\
-  -d '{"provider":"openai","api_key":"sk-...","model":"gpt-4o-mini"}'`}
+  -d '{"provider":"openai","api_key":"sk-...","model":"gpt-6-luna"}'`}
         className="my-4"
       />
 

web/messages/de/docs/monitor/ai-assistant.json

@@ -130,8 +130,8 @@
       "tagline": "Industriestandard. Sehr genau und weit verbreitet.",
       "items": [
         "API-Schlüssel: <a>platform.openai.com/api-keys</a>",
-        "Beispiele aus der mitgelieferten Modellliste: <code>gpt-4.1-nano</code>, <code>gpt-4.1-mini</code>, <code>gpt-4o-mini</code>, <code>gpt-4.1</code>, <code>gpt-4o</code>, <code>gpt-5-nano</code>.",
+        "Beispiele aus der mitgelieferten Modellliste: <code>gpt-6-luna</code>, <code>gpt-5.6-luna</code>, <code>gpt-5.4-mini</code>, <code>gpt-6-sol</code>, <code>gpt-5.5</code>.",
-        "Mitgelieferte Empfehlung: <strong><code>gpt-4.1-nano</code></strong>. Der Adapter unterstützt auch besondere Anforderungen von Reasoning-Modellen; die Liste enthält gpt-5-nano. Diese Modelle sind daher nicht pauschal ausgeschlossen."
+        "Mitgelieferte Empfehlung: <strong><code>gpt-6-luna</code></strong>, das günstigste aktuelle Modell. Alle Modelle der Liste sind Reasoning-Modelle; der Adapter sendet jedem die Parameter, die es akzeptiert."
       ],
       "baseUrlTitle": "OpenAI-kompatible Basis URL",
       "baseUrlBody": "Der OpenAI-Anbieter akzeptiert eine eigene <em>Basis-URL</em> für OpenAI-kompatible Endpunkte, etwa <strong>BytePlus / ByteDance (Kimi K2.5)</strong>, <strong>LocalAI</strong>, <strong>LM Studio</strong>, <strong>vLLM</strong>, <strong>Together AI</strong> und <strong>Fireworks AI</strong>, die das Format <code>/v1/chat/completions</code> unterstützen. Tragen Sie die URL im OpenAI-Bereich neben dem API-Schlüssel ein."

web/messages/en/docs/monitor/ai-assistant.json

@@ -130,8 +130,8 @@
       "tagline": "Industry standard. Very accurate and widely used.",
       "items": [
         "API key: <a>platform.openai.com/api-keys</a>",
-        "Examples from the bundled model list: <code>gpt-4.1-nano</code>, <code>gpt-4.1-mini</code>, <code>gpt-4o-mini</code>, <code>gpt-4.1</code>, <code>gpt-4o</code>, <code>gpt-5-nano</code>.",
+        "Examples from the bundled model list: <code>gpt-6-luna</code>, <code>gpt-5.6-luna</code>, <code>gpt-5.4-mini</code>, <code>gpt-6-sol</code>, <code>gpt-5.5</code>.",
-        "Bundled recommendation: <strong><code>gpt-4.1-nano</code></strong>. The adapter also handles model-specific requirements for reasoning models; the bundled list includes gpt-5-nano, so reasoning models are not universally excluded."
+        "Bundled recommendation: <strong><code>gpt-6-luna</code></strong>, the most economical current model. Every model in the bundled list is a reasoning model; the adapter sends each one the parameters it accepts."
       ],
       "baseUrlTitle": "OpenAI-compatible base URL",
       "baseUrlBody": "The OpenAI provider also accepts a custom <em>Base URL</em>, which lets you point the Monitor at any OpenAI-compatible endpoint. Confirmed to work with <strong>BytePlus / ByteDance (Kimi K2.5)</strong>, <strong>LocalAI</strong>, <strong>LM Studio</strong>, <strong>vLLM</strong>, <strong>Together AI</strong>, <strong>Fireworks AI</strong>, and any other service that speaks the OpenAI <code>/v1/chat/completions</code> dialect. Set the URL in the OpenAI tab next to the API key field."

web/messages/es/docs/monitor/ai-assistant.json

@@ -130,8 +130,8 @@
       "tagline": "Estándar de la industria. Muy preciso y ampliamente usado.",
       "items": [
         "Clave API: <a>platform.openai.com/api-keys</a>",
-        "Ejemplos de la lista de modelos incluida: <code>gpt-4.1-nano</code>, <code>gpt-4.1-mini</code>, <code>gpt-4o-mini</code>, <code>gpt-4.1</code>, <code>gpt-4o</code>, <code>gpt-5-nano</code>.",
+        "Ejemplos de la lista de modelos incluida: <code>gpt-6-luna</code>, <code>gpt-5.6-luna</code>, <code>gpt-5.4-mini</code>, <code>gpt-6-sol</code>, <code>gpt-5.5</code>.",
-        "Recomendación incluida: <strong><code>gpt-4.1-nano</code></strong>. El adaptador también gestiona requisitos específicos de los modelos de razonamiento; la lista incluye gpt-5-nano, por lo que no se excluyen todos esos modelos."
+        "Recomendación incluida: <strong><code>gpt-6-luna</code></strong>, el modelo actual más económico. Todos los modelos de la lista son de razonamiento; el adaptador envía a cada uno los parámetros que acepta."
       ],
       "baseUrlTitle": "Base URL compatible con OpenAI",
       "baseUrlBody": "El proveedor OpenAI también acepta una <em>Base URL</em> personalizada, que te permite apuntar el Monitor a cualquier endpoint compatible con OpenAI. Confirmado que funciona con <strong>BytePlus / ByteDance (Kimi K2.5)</strong>, <strong>LocalAI</strong>, <strong>LM Studio</strong>, <strong>vLLM</strong>, <strong>Together AI</strong>, <strong>Fireworks AI</strong> y cualquier otro servicio que hable el dialecto OpenAI <code>/v1/chat/completions</code>. Configura la URL en la pestaña OpenAI junto al campo de API key."

web/messages/fr/docs/monitor/ai-assistant.json

@@ -130,8 +130,8 @@
       "tagline": "Norme de l'industrie. Très précis et largement utilisé.",
       "items": [
         "Clé API : <a>platform.openai.com/api-keys</a>",
-        "Exemples de la liste de modèles fournie : <code>gpt-4.1-nano</code>, <code>gpt-4.1-mini</code>, <code>gpt-4o-mini</code>, <code>gpt-4.1</code>, <code>gpt-4o</code>, <code>gpt-5-nano</code>.",
+        "Exemples de la liste de modèles fournie : <code>gpt-6-luna</code>, <code>gpt-5.6-luna</code>, <code>gpt-5.4-mini</code>, <code>gpt-6-sol</code>, <code>gpt-5.5</code>.",
-        "Recommandation fournie : <strong><code>gpt-4.1-nano</code></strong>. L’adaptateur gère aussi les exigences des modèles de raisonnement ; la liste inclut gpt-5-nano, ils ne sont donc pas tous exclus."
+        "Recommandation fournie : <strong><code>gpt-6-luna</code></strong>, le modèle actuel le plus économique. Tous les modèles de la liste sont des modèles de raisonnement ; l’adaptateur envoie à chacun les paramètres qu’il accepte."
       ],
       "baseUrlTitle": "URL de base compatible OpenAI",
       "baseUrlBody": "Le fournisseur OpenAI accepte une <em>URL de base</em> personnalisée pour les points d’accès compatibles : <strong>BytePlus / ByteDance (Kimi K2.5)</strong>, <strong>LocalAI</strong>, <strong>LM Studio</strong>, <strong>vLLM</strong>, <strong>Together AI</strong>, <strong>Fireworks AI</strong> et les services utilisant le format <code>/v1/chat/completions</code>. Saisissez l’URL dans l’onglet OpenAI près de la clé API."

web/messages/it/docs/monitor/ai-assistant.json

@@ -130,8 +130,8 @@
       "tagline": "Standard di settore. Molto accurato e ampiamente usato.",
       "items": [
         "Chiave API: <a>platform.openai.com/api-keys</a>",
-        "Esempi dall’elenco di modelli incluso: <code>gpt-4.1-nano</code>, <code>gpt-4.1-mini</code>, <code>gpt-4o-mini</code>, <code>gpt-4.1</code>, <code>gpt-4o</code>, <code>gpt-5-nano</code>.",
+        "Esempi dall’elenco di modelli incluso: <code>gpt-6-luna</code>, <code>gpt-5.6-luna</code>, <code>gpt-5.4-mini</code>, <code>gpt-6-sol</code>, <code>gpt-5.5</code>.",
-        "Raccomandazione inclusa: <strong><code>gpt-4.1-nano</code></strong>. L’adattatore gestisce anche i requisiti dei modelli di ragionamento; la lista include gpt-5-nano, quindi questi modelli non sono esclusi in generale."
+        "Raccomandazione inclusa: <strong><code>gpt-6-luna</code></strong>, il modello attuale più economico. Tutti i modelli dell’elenco sono modelli di ragionamento; l’adattatore invia a ciascuno i parametri che accetta."
       ],
       "baseUrlTitle": "URL di base compatibile con OpenAI",
       "baseUrlBody": "Il provider OpenAI accetta un <em>URL di base</em> personalizzato per endpoint compatibili, come <strong>BytePlus / ByteDance (Kimi K2.5)</strong>, <strong>LocalAI</strong>, <strong>LM Studio</strong>, <strong>vLLM</strong>, <strong>Together AI</strong> e <strong>Fireworks AI</strong>, che usano il formato <code>/v1/chat/completions</code>. Imposta l’URL nella scheda OpenAI accanto alla chiave API."

web/messages/pt/docs/monitor/ai-assistant.json

@@ -130,8 +130,8 @@
       "tagline": "Padrão industrial. Muito preciso e amplamente utilizado.",
       "items": [
         "Chave API: <a>platform.openai.com/api-keys</a>",
-        "Exemplos da lista de modelos incluída: <code>gpt-4.1-nano</code>, <code>gpt-4.1-mini</code>, <code>gpt-4o-mini</code>, <code>gpt-4.1</code>, <code>gpt-4o</code>, <code>gpt-5-nano</code>.",
+        "Exemplos da lista de modelos incluída: <code>gpt-6-luna</code>, <code>gpt-5.6-luna</code>, <code>gpt-5.4-mini</code>, <code>gpt-6-sol</code>, <code>gpt-5.5</code>.",
-        "Recomendação incluída: <strong><code>gpt-4.1-nano</code></strong>. O adaptador também trata requisitos dos modelos de raciocínio; a lista inclui gpt-5-nano, pelo que esses modelos não são excluídos em geral."
+        "Recomendação incluída: <strong><code>gpt-6-luna</code></strong>, o modelo atual mais económico. Todos os modelos da lista são de raciocínio; o adaptador envia a cada um os parâmetros que aceita."
       ],
       "baseUrlTitle": "URL base compatível com OpenAI",
       "baseUrlBody": "O fornecedor OpenAI aceita um <em>URL base</em> personalizado para serviços compatíveis, como <strong>BytePlus / ByteDance (Kimi K2.5)</strong>, <strong>LocalAI</strong>, <strong>LM Studio</strong>, <strong>vLLM</strong>, <strong>Together AI</strong> e <strong>Fireworks AI</strong>, que utilizam o formato <code>/v1/chat/completions</code>. Defina o URL no separador OpenAI junto à chave API."

web/messages/sk/docs/monitor/ai-assistant.json

@@ -130,8 +130,8 @@
       "tagline": "Priemyselná norma. Veľmi presné a široko používané.",
       "items": [
         "Kľúč API: <a>platform.openai.com/api-keys</a>",
-        "Príklady z dodaného zoznamu modelov: <code>gpt-4.1-nano</code>, <code>gpt-4.1-mini</code>, <code>gpt-4o-mini</code>, <code>gpt-4.1</code>, <code>gpt-4o</code>, <code>gpt-5-nano</code>.",
+        "Príklady z dodaného zoznamu modelov: <code>gpt-6-luna</code>, <code>gpt-5.6-luna</code>, <code>gpt-5.4-mini</code>, <code>gpt-6-sol</code>, <code>gpt-5.5</code>.",
-        "Dodané odporúčanie: <strong><code>gpt-4.1-nano</code></strong>. Adaptér spracúva aj osobitné požiadavky modelov uvažovania; zoznam obsahuje gpt-5-nano, takže tieto modely nie sú všeobecne vylúčené."
+        "Dodané odporúčanie: <strong><code>gpt-6-luna</code></strong>, najúspornejší aktuálny model. Všetky modely v zozname sú modely uvažovania; adaptér každému posiela parametre, ktoré prijíma."
       ],
       "baseUrlTitle": "Základné URL kompatibilné s OpenAI",
       "baseUrlBody": "Poskytovateľ OpenAI prijíma vlastnú <em>Základnú URL</em> pre kompatibilné služby, napríklad <strong>BytePlus / ByteDance (Kimi K2.5)</strong>, <strong>LocalAI</strong>, <strong>LM Studio</strong>, <strong>vLLM</strong>, <strong>Together AI</strong> a <strong>Fireworks AI</strong>, ktoré používajú formát <code>/v1/chat/completions</code>. URL nastavte na karte OpenAI vedľa kľúča API."

web/messages/sv/docs/monitor/ai-assistant.json

@@ -130,8 +130,8 @@
       "tagline": "Industristandard. Mycket exakt och allmänt används.",
       "items": [
         "API-nyckel: <a>platform.openai.com/api-keys</a>",
-        "Exempel från den medföljande modellistan: <code>gpt-4.1-nano</code>, <code>gpt-4.1-mini</code>, <code>gpt-4o-mini</code>, <code>gpt-4.1</code>, <code>gpt-4o</code>, <code>gpt-5-nano</code>.",
+        "Exempel från den medföljande modellistan: <code>gpt-6-luna</code>, <code>gpt-5.6-luna</code>, <code>gpt-5.4-mini</code>, <code>gpt-6-sol</code>, <code>gpt-5.5</code>.",
-        "Medföljande rekommendation: <strong><code>gpt-4.1-nano</code></strong>. Adaptern hanterar även särskilda krav för resonemangsmodeller; listan innehåller gpt-5-nano, så sådana modeller är inte generellt uteslutna."
+        "Medföljande rekommendation: <strong><code>gpt-6-luna</code></strong>, den mest ekonomiska aktuella modellen. Alla modeller i listan är resonemangsmodeller; adaptern skickar varje modell de parametrar den accepterar."
       ],
       "baseUrlTitle": "OpenAI-kompatibel bas URL",
       "baseUrlBody": "OpenAI-leverantören accepterar en egen <em>Bas-URL</em> för kompatibla tjänster som <strong>BytePlus / ByteDance (Kimi K2.5)</strong>, <strong>LocalAI</strong>, <strong>LM Studio</strong>, <strong>vLLM</strong>, <strong>Together AI</strong> och <strong>Fireworks AI</strong>, som stöder formatet <code>/v1/chat/completions</code>. Ange URL:en på OpenAI-fliken bredvid API-nyckeln."