To continue improving model quality and service reliability, the following models will be retired on August 12, 2026. After retirement, these models will stop serving inference requests, and applications that still call them will no longer receive valid responses.
Please complete your model replacement, parameter tuning, and full end-to-end validation before August 12, 2026 to keep your production services running.
Retired models and recommended replacements
The table below lists the models being retired and the recommended replacement for each. In most cases, migrating means replacing the model field in your request with the replacement model ID.
step-tts-2 and stepaudio-2.5-tts use different control paradigms: step-tts-2 is tag-based (preset emotions and styles), while stepaudio-2.5-tts is controlled with natural-language context. Review how you specify delivery instead of only swapping the model ID. If you use cloned voices created with step-tts-2, validate them against stepaudio-2.5-tts and re-clone them if needed — see Voice cloning.How to migrate
- Find the model you currently use in the table above and note its recommended replacement.
- Replace the
model field value in your API requests with the replacement model ID — for example, change step-tts-2 to stepaudio-2.5-tts.
- Review the replacement model’s capabilities on its model page (linked above) and its rates on Pricing & Rate Limits. We recommend running full validation before switching production traffic.
If you have questions, see the FAQ. 
