Inworld models, access and benchmark guide
Inworld model and access guide with a practical US-English evaluation brief and links to publisher materials.

👁 111 reads
For a game character, speech synthesis produces a line of dialogue. A full conversation also needs listening and reasoning. Inworld lists TTS rates separately from the other components, so a script’s character count cannot price an entire live session.
Choosing the first job
Have an invented station attendant repeat a corrected destination after an interruption; inspect the saved transcript.
Models in the November 17 snapshot
| Model and track | Placement | Reported score |
|---|---|---|
| Realtime TTS-2 (AA provider voice) | 9 | 1270 Elo |
| Realtime TTS-2 (AA controlled voice) | 9 | 1145 Elo |
| Realtime TTS-2 Flash (AA provider voice) | 18 | 1215 Elo |
| Realtime TTS-2 Flash (AA controlled voice) | 20 | 1066 Elo |
| Inworld Realtime (Default Cascaded System) (AA live voice) | 25 | 884 Elo |
Release record. Publisher documentation identifies the product family. The ranked configurations include Realtime TTS-2, Realtime TTS-2 Flash, Inworld Realtime (Default Cascaded System).
Google · Your sourcesQuantslant higher in your Google - one clickAdd to preferred sources →Sources
- https://docs.inworld.ai/ (accessed November 17, 2026)
- https://inworld.ai/tts (accessed November 17, 2026)
- https://artificialanalysis.ai/text-to-speech/leaderboard/provider-voice (accessed November 17, 2026)
- https://artificialanalysis.ai/text-to-speech/leaderboard/controlled-voice (accessed November 17, 2026)
- https://artificialanalysis.ai/speech-to-speech/arena (accessed November 17, 2026)
More on this from QUANTSLANT
video·Image-to-video rankings for a shot with a first frameWhich systems can animate a supplied frame? Compare image-to-video rankings and plan a continuity check before paying for rerenders.