Google Releases Gemini 3.8 Flash TTS for General Availability
Google announced the general availability of Gemini 3.8 Flash TTS on September 22, 2026.
Official Announcement Details
Google officially released Gemini 3.8 Flash TTS as generally available on September 22, 2026. This release coincided with the launch of the Gemini 3.8 Flash-Lite TTS model and the new Gemini API Voices endpoint located at /v1beta/voices. The source event date is distinct from this article's publication date, which reflects current reporting on that specific historical announcement.
- Model identifier: gemini-3.8-flash-tts
- Provider: Google
- Input modality: Text
- Output modality: Audio
Documented Capabilities and Limits
The official changelog describes Gemini 3.8 Flash TTS as a flagship creative text-to-speech model. It is engineered for studio-grade voice fidelity, nuanced acting, regional dialects, and long-form multi-turn stability. The model has a documented context window of 8,192 tokens and a maximum output of 16,384 tokens. Open weights are not confirmed for this release. Pricing information is not provided in the source facts. No architectural details or local installation requirements are specified in the available documentation.
Evaluation Procedure for Engineers
Engineers can evaluate the model by testing its performance against specific criteria defined in the release notes. First, test the model’s ability to handle regional dialects by inputting text containing varied phonetic structures. Measure the audio output for clarity and consistency across different dialect samples. Compare the results against previous baseline recordings if available to assess fidelity. Validate long-form stability by generating multi-turn conversations exceeding standard short-form lengths to check for degradation in voice quality over time. These steps allow for a practical assessment of the documented features without assuming unverified benefits.
Primary sources
Frequently asked questions
What is the context window for Gemini 3.8 Flash TTS?
The model has a context window of 8,192 tokens and a maximum output limit of 16,384 tokens.
Are open weights available for this model?
Open weights are not confirmed in the source facts. The documentation does not establish whether local installation is possible or if API access is mandatory.