v0.11.0

Added

• OpenAI realtime, audio, speech and transcription models. gpt-realtime-2.1, gpt-realtime-2.1-mini,
  gpt-realtime-2, gpt-realtime-1.5, gpt-realtime-mini, gpt-realtime (text rate plus audio and image input,
  audio and image cache-read and audio output as modality prices), gpt-audio-1.5, gpt-audio-mini, gpt-audio,
  gpt-4o-mini-tts (TextToSpeech, output = audio), gpt-4o-transcribe, gpt-4o-mini-transcribe and
  gpt-4o-transcribe-diarize (SpeechToText), from OpenAI's pricing and model pages (2026-10-10).

Changed

• Breaking: a speech, transcription, search, realtime or image id no longer resolves to the chat model whose name it
  contains. gpt-4o-audio-preview, gpt-4o-realtime-preview, gpt-4o-mini-tts, gpt-4o-transcribe and
  gpt-4o-search-preview matched gpt-4o / gpt-4o-mini through their contains aliases and were priced as chat
  output. A fuzzy match is now not taken when the rest of the id names a variant (audio, realtime, tts, transcribe,
  search, image, embedding, live, translate, moderation); such an id resolves only by a row or alias of its own
  and is otherwise unknown (null). Migration: register those models with CostCalculator.RegisterModel if you price them.

Fixed

• The catalog loads in applications that disable reflection-based JSON (Native AOT, trimmed or file-based apps):
  ModelCatalog threw TypeInitializationException there. The embedded catalog is now read with source-generated
  metadata, and the package is marked IsAotCompatible.