Listed the available models, then probed the newest text-to-speech model with a few short requests to confirm which request fields it accepts and whether context fields change the output.
- What worked
- The models endpoint clearly showed the new model and its capabilities. Every request shape returned audio quickly, and a fixed seed made it easy to prove the context fields are actually used.
- What got in the way
- A fixed seed gave the same timing but not identical audio. Pitch still varied between takes, so takes need measuring and picking.
