Breaking

Inworld AI Releases Realtime TTS-2 and Realtime TTS-2 Flash

September 2, 2026 at 21:03 EDT

Media Generation
Foundation Models

Inworld AI has released Realtime TTS-2, an expressive voice model for live conversation, alongside Realtime TTS-2 Flash, a lower-latency, lower-cost variant aimed at latency-sensitive and high-volume applications. Inworld says Realtime TTS-2 entered general availability on August 31, 2026 through the Inworld API and Inworld Realtime API.

31 AUG 2026 · INWORLD AI

Realtime voice gets two lanes

Realtime TTS-2 prioritizes expressive control for conversation. Realtime TTS-2 Flash targets lower latency and cost for high-volume output.

Server-side P99 time to first byte

under 100ms
Realtime TTS-2

25ms
Realtime TTS-2 Flash

The figures are server-side P99 measurements; network round trips are not included.

5–15 sec

Audio for instant cloning

200+

Languages listed for both models

$15 / 1M

Flash on-demand rate

Where TTS-2 fits

Natural-language steering
Inline non-verbal cues
Expressive conversational delivery

Where Flash fits

Latency-critical applications
High-volume workloads
Steering tags may be read literally

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.