Google announced the launch of Gemini 3.8 Flash TTS and Gemini 3.8 Flash-Lite TTS, its most expressive audio models yet, supporting custom voices across 100+ languages and 2,000+ ready-to-use voices. The models are rolling out starting today to developers via Google AI Studio and the Gemini API, and to consumers via Gemini Notebook and Google Vids, with Gemini Enterprise availability coming soon.
SynthePulse Insight · AI deep reading
Version 1 · 1 source
Google announced via its official X account the launch of Gemini 3.8 Flash TTS and Gemini 3.8 Flash-Lite TTS, emphasizing expressiveness, multilingual support and line-by-line control, along with a phased rollout plan for developers and consumers. But all capabilities and figures come from the vendor's own statements; the source contains no third-party independent verification, and it lacks verifiable metrics such as pricing, latency and audio quality benchmarks.
The official Google AI X account posted on September 23, 2026, announcing the launch of two text-to-speech models, Gemini 3.8 Flash TTS and Gemini 3.8 Flash-Lite TTS. The post mentioned a companion article from blog.google titled "Gemini 3.8 text-to-speech says hello," but the given source does not provide the content of that article, so the details cannot be cross-checked.
In terms of dissemination data, the post shows about 229,600 views, along with 135 replies, 209 reposts, 2.1K likes and 845 bookmarks. These figures indicate the announcement received some attention, but attention itself does not constitute verification of the models' capabilities.
It must be made clear that the only source this article relies on is Google's official release content. Apart from that official post, there is no third-party independent verification of the above capabilities or figures. Therefore, the capabilities, language counts, voice counts and audio duration mentioned below are all vendor claims, not independently verified conclusions.
Google claims these are its most expressive audio models to date. This statement is a vendor self-assessment; the source provides no quantifiable evaluation criteria and no third-party comparison data, so it is impossible to judge which models "most expressive" is being measured against, or on what tasks it holds.
On coverage, Google says users can create custom voices across 100+ languages, and says it offers 2,000+ ready-to-use voices to choose from. These two figures are the most specific quantitative metrics in this release, but the source does not explain the specific list, the quality of coverage, or whether all these voices are actually available.
In other words, 100+ languages and 2,000+ voices can currently only be understood as Google's claims. Whether they correspond to a complete, actually callable capability set will need to be confirmed through official documentation or third-party testing.
Google says the model can direct back-and-forth dialogue and control the expression of speech line by line. This means users are not just inputting a block of text and waiting for synthesis; they can intervene in the direction of the dialogue and the expression of each line.
Google also says natural cues like <laughs> can be added, or active-listening interjections like |mhm|. If these cues and interjections work as claimed, they would directly affect the naturalness and conversational feel of the synthesized speech.
On audio duration, Google says it can generate hours of consistent, glitch-free audio. This statement likewise lacks quantifiable evaluation criteria and third-party comparison. The actual effects and supported scope of line-by-line control, dialogue direction, and cues such as <laughs> and |mhm| are all unverified in the source.
Google positions Gemini 3.8 Flash TTS as a model that can design custom voice personas from scratch and offers line-level control, targeting high-fidelity creative production such as games, immersive audiobooks and podcasts. This positioning corresponds to scenarios with high demands on voice persona and expressive detail.
Google positions Gemini 3.8 Flash-Lite TTS as an engine that can automatically adjust tone and pacing in real time, targeting near-real-time voice agents, low-cost scaling, high-volume dubbing and batch audio creation. This positioning corresponds to scenarios more sensitive to cost and throughput, with relatively lower requirements for fine line-by-line control.
The two models share the same generation name, but their positioning differs markedly: one emphasizes designing voice personas from scratch and line-by-line control, the other emphasizes automatically adjusting tone and pacing, targeting near-real-time use and scale. This division of labor itself is Google's official positioning, not a third-party review conclusion.
Google says that starting on launch day, developers can use both models in Google AI Studio and the Gemini API. This is the most time-specific rollout information in this release, corresponding to immediate availability on the developer side.
On the consumer side, Google says consumers can use Gemini 3.8 Flash TTS in Gemini Notebook and Gemini 3.8 Flash-Lite TTS in Google Vids. That is, the two models do not land in the same place among consumer products; they correspond to different product scenarios.
Google also says both models are coming soon to Gemini Enterprise. But "coming soon" gives no clear timetable, and the specific launch time and form are unknown. In addition, the source does not state whether and when the models will be available in regions such as mainland China.
The source is a social media post and does not provide verifiable technical metrics such as pricing, latency or audio quality benchmarks. For text-to-speech models, these metrics often directly determine whether they can enter production environments, but this release does not address them.
The pricing, rate limits and commercial licensing terms of the two models are also not explained in the source. For teams planning to use them for high-volume dubbing, voice agents or commercial content production, the absence of this information means cost and compliance assessments cannot be completed.
In addition, the specific list of 100+ languages and 2,000+ voices, the quality of coverage and whether all are available remain unclear; the specific timing and form of the Gemini Enterprise launch are unknown; and the content of the blog.google companion article mentioned in the post is not provided in the source, so the details cannot be cross-checked.
All facts in this article come from the post published by the official Google AI X account on September 23, 2026, and the information attached to it. That post is an official vendor release, and its statements about model capabilities, language counts, voice counts, audio duration and controllability are all Google's own claims; the source contains no third-party independent verification. The content of the blog.google companion article mentioned in the post is not provided in the source and cannot be cross-checked. Therefore, all capabilities and figures in the article should be understood as vendor claims, not independently verified conclusions; where statements such as "most expressive" and "hours of consistent, glitch-free audio" appear, this article preserves their source attribution.
Google used a single official post to simultaneously release Gemini 3.8 Flash TTS and Gemini 3.8 Flash-Lite TTS, and laid out three rollout paths for developers, consumers and enterprise. What is truly worth noting is not self-assessments like "most expressive," but the division of labor between the two models—high-fidelity creative production versus low-cost scaling—and the information gaps exposed by the phased rollout: pricing, latency, audio quality benchmarks, commercial licensing and regional availability are all unstated in the source. Until these questions are answered, this release is better seen as a vendor roadmap than as a verifiable capability list.
Primary report
Primary source