I’m wondering if there is a way to maintain cached tokens between gpt-realtime sessions. For something like instructions or a large system message that is the same for each realtime session, can a new session take advantage of cached text tokens from a previous session to cut costs?
Related topics
| Topic | Replies | Views | Activity | |
|---|---|---|---|---|
| Charging for adding history to realtime API | 1 | 187 | October 2, 2024 | |
| (Realtime API) Hows everyone managing longer than 30min sessions | 4 | 1405 | June 9, 2025 | |
| Realtime API Prompt Caching | 1 | 695 | October 31, 2024 | |
| Reducing cost of Realtime API by caching tool reaponses | 2 | 548 | October 5, 2024 | |
| Cached input audio_tokens is always 0 | 3 | 506 | November 8, 2024 |