LLMs: OpenAI: adopt uncurated marker for native unknowns

Author: enricoros

src/modules/llms/server/models.mappings.ts

@@ -20,7 +20,8 @@ import type { ModelDescriptionSchema } from './llm.server.types';
  * Readers: llm-registry-sync (marked + null contextWindow = held off the publication push),
  *          listModels.test, the website (strips brackets, sinks pubDate; own regex, keep in sync).
  *
- * Legacy, not unified: bedrock ' [?]' suffix, openai bare '?'.
+ * Legacy, not unified: bedrock ' [?]' suffix; openai-compat hosts keep a lenient bare '?'
+ * (unknown is the norm there, marking would blank whole services from publication).
  */
 const LLM_LABEL_UNCURATED = '[?]';
 

src/modules/llms/server/openai/models/openai.models.ts

@@ -1200,26 +1200,38 @@ export const _knownOpenAIChatModels = llmsDefineModels<_OpenAIModelDef>()([
 // -- 0-day or unknown models --
 
 export function llmsFallbackForOpenAIModel(modelId: string, isLikelyOpenAI: boolean): KnownModel {
-  return isLikelyOpenAI && _isLikelyResponsesAPIModel(modelId) ? _llmsOpenAIModelResponsesFallback : _llmsOpenAIModelFallback;
+  if (!isLikelyOpenAI)
+    return _llmsOpenAICompatModelFallback;
+  return _isLikelyResponsesAPIModel(modelId) ? _llmsOpenAIModelResponsesFallback : _llmsOpenAIModelFallback;
 }
 
+// native OpenAI (and Azure deployments): the ids-only list discloses no type or context, so
+// unknowns carry the uncurated marker (see llmsLabelUncurated) and a null (never guessed) context
 const _llmsOpenAIModelFallback: KnownModel = {
   idPrefix: '',
-  label: '?',
-  description: 'Unknown, please let us know the ID. Assuming a context window of 128k tokens, and a maximum output of 4k tokens.',
-  contextWindow: 128000,
-  maxCompletionTokens: 4096,
+  label: '[?]', // + ' [<id>]' via the variant append below
+  description: 'Unknown OpenAI model, not yet curated - please let us know the ID.',
+  contextWindow: null,
   interfaces: IFS_CHAT_MIN,
   // hidden: true,
 };
 
 const _llmsOpenAIModelResponsesFallback: KnownModel = {
   ..._llmsOpenAIModelFallback,
-  description: 'Unknown OpenAI model. Assuming Responses API support, 128k context, 16k output.',
-  maxCompletionTokens: 16384,
+  description: 'Unknown OpenAI model, not yet curated - assuming Responses API support.',
   interfaces: [LLM_IF_OAI_Responses, ...IFS_CHAT_MIN],
 };
 
+// OpenAI-compatible hosts: unknown is the norm there, so stay lenient and unmarked
+// ('[?]' would mark every model and hold whole services off the registry-sync push)
+const _llmsOpenAICompatModelFallback: KnownModel = {
+  ..._llmsOpenAIModelFallback,
+  label: '?', // + ' [<id>]'
+  description: 'Unknown, please let us know the ID. Assuming a context window of 128k tokens, and a maximum output of 8k tokens.',
+  contextWindow: 128000,
+  maxCompletionTokens: 8192,
+};
+
 function _isLikelyResponsesAPIModel(modelId: string): boolean {
   // GPT-5+ family (gpt-5, gpt-5.1, gpt-5.2, gpt-6, etc.)
   if (/^gpt-[5-9]/.test(modelId)) return true;