Inworld has officially released its Realtime TTS-2 text-to-speech model for general availability. Currently ranking #1 on Artificial Analysis' Controlled Voice Arena, it is touted as the fastest in its class, with improvements driven by user data from its earlier research preview phase.
SynthePulse Insight · AI deep reading
Version 1 · 1 source
On September 2, 2026, Inworld AI announced the general availability (GA) of its text-to-speech model, Realtime TTS-2. The company claims the model currently ranks first in Artificial Analysis's 'Controlled Voice Arena' and is dubbed the fastest TTS in its class. Its core feature allows developers to embed delivery instructions ranging from short emotion tags to full tone descriptions directly into text.
Inworld AI announced via social media on September 2, 2026, that its real-time text-to-speech model, Realtime TTS-2, has entered the General Availability (GA) phase. This release marks a critical transition for the product from the research preview stage to commercial deployment.
Inworld officially claims that Realtime TTS-2 currently holds the number one spot in the 'Controlled Voice Arena' on the Artificial Analysis platform. Simultaneously, the company defines it as the 'fastest TTS in its class.' It is important to note that these rankings and speed comparisons are based on data from Artificial Analysis and Inworld's own statements; specific benchmark testing details were not elaborated upon in the source material.
A significant technical feature of this model is the ability for developers to write 'delivery instructions' directly into the text. This control mechanism offers high flexibility, covering a range from short emotion tags to complete sentence descriptions used to specify the intonation of the voice output. This design aims to address the difficulties in controlling intonation in traditional speech synthesis, reportedly significantly boosting development efficiency.
Inworld pointed out that the research preview version of Realtime TTS-2 received usage volumes far exceeding expectations. The company integrated the usage experiences of these early users back into its research, resulting in the GA version released now. Inworld describes this version as the 'biggest leap' in its history, implying substantial improvements in performance or functionality compared to previous generations.
This report is primarily based on press releases published by Inworld AI and its partners on the X platform. Claims regarding 'ranking #1 in Artificial Analysis' and being the 'fastest TTS' are classified as source claims, lacking independent third-party verification data. Details regarding the model's internal architecture or specific latency in milliseconds were not provided in the source.
By emphasizing fine-grained textual control over voice intonation and its high ranking in Artificial Analysis, Inworld attempts to establish Realtime TTS-2's leading position in the competitive TTS market. Developers should focus on the practical effectiveness of its instruction controls rather than relying solely on vendor performance claims.
Primary report
Primary source