Gemini 3.5 Live Translate Brings Low-Latency Speech Translation to Apps and Meet
Google released Gemini 3.5 Live Translate for near real-time speech-to-speech translation across 70+ languages, with developer preview access through the Gemini Live API.
Google released Gemini 3.5 Live Translate on June 9, 2026 as an audio model for near real-time speech-to-speech translation. Google says it supports more than 70 languages, streams translated speech continuously instead of waiting for full turns, and is rolling out through the Gemini Live API, Google AI Studio, Google Translate, and a private preview for Google Meet. The DeepMind model card adds implementation details: audio input can use up to a 128K token context window, while output can include audio and text.
Key takeaways
- Gemini 3.5 Live Translate is aimed at live speech-to-speech translation, not static document translation.
- Google says the model preserves intonation, pacing, and pitch while staying a few seconds behind the speaker during live sessions.
- Developers can try it in public preview through the Gemini Live API and Google AI Studio.
- Google Meet support starts with select business Google Workspace customers in private preview before a broader rollout.
- The model card lists limitations around voice consistency, accents, rapid language switching, background noise, and multi-speaker sessions.
Practical LinkLoot angle
This release is useful for teams building multilingual calls, creator dubbing workflows, support desks, live lessons, and field communication tools. The important shift is that translation becomes a streaming audio component an app can call, not a separate transcription-then-translation-then-TTS pipeline. That can reduce handoff latency and simplify product design, but it does not remove the need for consent, caption fallback, and human review in regulated or high-stakes contexts.
| Use case | Best fit | Watch out for | Source |
|---|---|---|---|
| Live customer support | Low-latency spoken translation inside calls | Accents, noisy environments, and support-script accuracy | Google Blog, model card |
| Multilingual meetings | Google Meet speech translation for business users | Private preview availability and Workspace eligibility | Google Blog |
| Developer apps | Gemini Live API and AI Studio prototypes | Preview status, pricing, rate limits, and safety terms | Google AI docs |
| Creator translation | Dubbing or live event translation | Voice drift, speaker changes, and disclosure requirements | Model card |
What to verify before you act
Check availability first: public preview for developers is not the same as production readiness, and Meet access starts as a private preview for selected Workspace customers. Review the model card before using it for live events, because the listed limits affect exactly the scenarios that make speech translation hard: non-native accents, similar languages, rapid switches, pauses, background audio, and multiple speakers. If translated audio is published or rebroadcast, confirm disclosure and watermarking requirements; Google says generated audio is watermarked with SynthID.
Source check
- Google's announcement confirms the June 9 release, 70+ language support, Google Translate rollout, Gemini Live API public preview, and Meet private preview.
- The Google DeepMind model card confirms model scope, input/output shape, evaluation dimensions, safety review framing, and known limitations.
- The Google AI developer documentation confirms the
gemini-3.5-live-translate-previewmodel path and low-latency audio-to-audio positioning. - Product Hunt independently lists Gemini 3.5 Live Translate as a launch item focused on live speech-to-speech translation.
For teams turning this into a repeatable workflow, LinkLoot's AI workflow automation guide is the better next stop than a generic model roundup.
It is Google's audio-to-audio model for low-latency live speech translation across more than 70 languages.