Is OpenAI actively making changes on gpt-4o-mini-tts-2025-12-15?
The speech generated today on 2/10 is noticeably worse than output from 1/13, even though I used the exact same snapshot.
The audio sounds much darker, almost as if an aggressive low-pass filter was applied, resulting in reduced clarity and degraded vocal tone.
For example, Shimmer and Nova now sound very similar, whereas they were clearly distinct when I generated earlier.
I had to go back to gpt-4o-mini-tts-2025-03-20.
Are there other snapshots?
Hey @tiffiana, that’s odd. To help us understand what’s happening, could you please share the time window (including time zone), numbered steps to reproduce with expected vs actual behavior, the exact error text (copy/paste) plus a sample of the audio if possible, your environment details (browser/app name and version, OS and version, and device model if mobile), and the request ID/trace ID if available.
please ping us at support@openai.com and share the above information there so we can look into this further.
i have a similar experience the audios i generated a month ago sounded more vibrant and alive. now they sound monoton and dull, like it is ignoring my voice instructions. any idea?
I’ve noticed the same thing. The audio from before felt more lively, but now it sounds flatter and a bit monotone, like it’s not really picking up on the voice instructions.
And soon they will remove the much better model
well, I will not be using the gpt-4o-mini-tts-2025-12-15 because its just awful in every way of how it sounds, compared to gpt-4o-mini-tts-2025-03-20
This will be a sad day, the gpt-4o-mini-tts-2025-03-20 Shimmer was perfect
My experience
Not only has the quality gotten worse, but they’re also removing the model that followed instructions the best, and the official “replacement” is GPT Realtime, which doesn’t make much sense because it doesn’t work the same way and completely breaks older code.