Model comparison
Compare GPT-Realtime-2.1 and Inworld Realtime TTS-2 using the same provider-sourced voice & realtime rubric. No mystery score and no invented benchmark ranking.
Facts checked September 4, 2026
Quick take
OpenAI's realtime speech-to-speech reasoning model for tool-using voice agents that also need text and image context.
There is no published universal latency number, and audio-token costs are hard to compare directly with minute- or character-priced competitors.
Inworld's newest expressive realtime text-to-speech model, designed to remember conversational delivery and speak in more than 100 languages.
This is the speech layer, not a complete agent. Inworld's current zero-data-retention page does not explicitly list TTS-2, so confirm coverage before sending sensitive text or voice data.
Compare the published facts
Values use each provider's own published units and limits. A blank means the provider did not publish a directly comparable value in the sources reviewed.
| Voice & realtime | GPT-Realtime-2.1 | Inworld Realtime TTS-2 |
|---|---|---|
| Price basisToken-, character-, or minute-based price for the listed access route. | Audio $32 in · $64 out / 1M tokens | $25 / 1M chars on demand; volume discounts |
| Interaction modeSpeech-to-speech, audio-to-audio, or streamed text-to-speech behavior. | Speech-to-speech with text/image context | Steerable realtime TTS with prior-audio context |
| Published latencyProvider-published measurement with its stated exclusions; not a Cody test. | Not published | Under 200ms median TTFA |
| LanguagesProvider-documented language coverage or a clear unpublished marker. | Multilingual; exact list not published here | 100+ with mid-utterance switching |
| Tools & controlNative function calling, reasoning, or speech-control features. | Function calling and configurable reasoning | Voice direction, design, cloning, non-verbals |
| Context or input limitDocumented token or character limit where it is meaningful. | 128K input · 32K max output | Not published |
How to choose
Start with the job you need to complete, then validate cost, access, and policy details on your exact provider route.
OpenAI says API content is not used for training by default. Default abuse-monitoring logs may be retained up to 30 days, with additional controls available to qualifying organizations.
Provider and API links
Inworld advertises zero data retention as an enterprise add-on, but its currently published ZDR support page names only TTS-1.5 Mini and Max. Confirm TTS-2 coverage directly before sending regulated or confidential text.
Frequently asked questions
GPT-Realtime-2.1: OpenAI's realtime speech-to-speech reasoning model for tool-using voice agents that also need text and image context. Inworld Realtime TTS-2: Inworld's newest expressive realtime text-to-speech model, designed to remember conversational delivery and speak in more than 100 languages.
Consider GPT-Realtime-2.1 when your priority is Tool-using customer or employee voice agents. Consider Inworld Realtime TTS-2 when your priority is Realtime companions and support voices. Test both with your own data and provider route before committing.
No. This comparison aligns provider-published facts for the Voice & realtime category. It does not claim a universal winner or combine incompatible third-party benchmark scores.
