The input audio tokens a run stored that were not cache hits: Input.AudioTokens less Input.CachedAudioTokens,
never below 0. Providers count cached audio in the audio total (OpenAI's cached_tokens_details.audio_tokens is a
subset of audio_tokens), and cached audio is priced at the cache-read rate, not the audio rate.
The input audio tokens a run stored that were not cache hits:
Input.AudioTokenslessInput.CachedAudioTokens, never below 0. Providers count cached audio in the audio total (OpenAI'scached_tokens_details.audio_tokensis a subset ofaudio_tokens), and cached audio is priced at the cache-read rate, not the audio rate.