No new or removed documentation pages in this update.
Deprecations (/gemini-api/docs/deprecations)
- Removed
gemini-3.1-flash-tts-previewfrom the preview models table and now recommendsgemini-3.8-flash-ttsorgemini-3.8-flash-lite-ttsas replacements for the Gemini 2.5 TTS preview models.
Models (/gemini-api/docs/models)
- The catalog now lists Gemini 3.8 Flash TTS and Gemini 3.8 Flash-Lite TTS, marks Gemini 3.1 Flash TTS as Preview, and describes
gemini-3.1-flash-tts-previewas a legacy TTS preview model with a migration note to the 3.8 TTS models. - The Gemini 3.8 TTS model pages now describe language coverage as over 130 / over 100 languages, label the 16,384 output token figure as a Gemini API serving limit, and date the latest update as September 2026; the Flash TTS page no longer documents inline IPA overrides.
Speech generation (/gemini-api/docs/speech-generation)
- Page descriptions now call out
gemini-3.8-flash-ttsandgemini-3.8-flash-lite-tts; language tables add Turkish and Vietnamese and use “over 130” / “over 100” coverage; REST examples show how to extract base64 audio to a WAV file; and multi-speaker concatenation guidance on/gemini-api/docs/generate-content/speech-generationnow covers unaryaudio/wavRIFF headers and requesting raw PCM (AUDIO_L16).
Voice design and replication
- REST examples on
/gemini-api/docs/voice-designand/gemini-api/docs/voice-replicationnow pipe the last model audio payload throughjqand base64-decode it toout.wav.
Coding agents (/gemini-api/docs/coding-agents)
- Documents the no-MCP fallback as
/gemini-api/docs/llms.txt, including fetching individual pages as raw Markdown with a.md.txtsuffix (for examplehttps://ai.google.dev/gemini-api/docs/speech-generation.md.txt).