The performance on the TTS models have improved greatly the last couple of months, and I for one am impressed. However the token-cost still remain quite high, and for production settings it dominates the total cost of certain implementations.
Other models have batch-pricing through batch-api and flex operations. Implementing the same for the TTS models (particularily 4o-mini-tts) would be greatly appreciated. A more lightweight TTS model could also improve this, further enabling TTS integration in more areas.