Transcribes audio into the input language using a unified OpenAI-compatible multipart interface.
Supports non-streaming JSON responses and, for capable OpenAI models, SSE streaming.

Required fields (this endpoint):

Provider-native requirements (mapped via model prefix; see linked docs for full rules):

On this endpoint always send file + model. ElevenLabs URL-based input (source_url) is only available
when the platform exposes it for elevenlabs/* models.

Optional parameters depend on the provider. Common examples:

• OpenAI — language, prompt, response_format (json, text, srt, verbose_json, vtt, diarized_json),
  temperature, timestamp_granularities[], stream, include[], chunking_strategy, known_speaker_names[],
  known_speaker_references[] (diarization models).
• ElevenLabs — language_code, diarize, num_speakers, timestamps_granularity, tag_audio_events,
  use_multi_channel, keyterms, webhook, source_url, entity_detection, temperature, seed, and others
  listed in ElevenLabs Speech-to-Text.

The set of accepted optional fields and response fields may differ between providers and models.
Use the model/info endpoint to check capabilities (e.g. streaming, diarization, word timestamps).