Gemini 3.5 Live Translate — Gemini Live Translation¶

Development line: project:gemini-3-5-live-translate · thread gemini-live-translation
Last event: 2026-06-13 · 1 dated since 2026-06-13 · Researched: 2026-09-05 · confidence: high

What it is¶

Gemini 3.5 Live Translate is Google’s low-latency audio-to-audio translation model for real-time spoken conversations.

• Automatic detection across 70+ languages
• Continuous translated speech
• Optional input and output transcripts

The model remains in preview, accepts audio only, and does not support tools or instructions. Use it as an interpreter pipeline, not as a general-purpose voice agent.

Development line¶

• 2026-06-13 — Google presented Gemini 3.5 Live Translate preview. On 2026-06-13, Google introduced Gemini 3.5 Live Translate as a preview model for real-time translation in Gemini Live. The release provided a Live API guide, an AI Studio Live preview URL, and integration references for LiveKit and Pipecat.

What changed¶

• 2026-06-13 — Gemini 3.5 Live Translate became available to developers in public preview through the Gemini Live API and Google AI Studio.
• 2026-08-26 — Google published the Gemini 3.5 Audio model card. It documented Live Translate’s 128K input and 64K output context limits, along with failure modes for voice, language detection, and background noise.

How to use this¶

From 2026-06-13, evaluate Gemini 3.5 Live Translate preview through the Gemini Live API or AI Studio. Confirm the LiveKit or Pipecat integration path before adopting it.

1. Create a Gemini API client and open a Live API session with model gemini-3.5-live-translate-preview. — https://ai.google.dev/gemini-api/docs/live-api/live-translate
2. Set response_modalities to audio and provide translationConfig.targetLanguageCode as a BCP-47 target-language code. — https://ai.google.dev/gemini-api/docs/live-api/live-translate
3. Stream 16-bit, 16 kHz, mono, little-endian PCM in roughly 100 ms chunks; consume the returned 24 kHz PCM audio. — https://ai.google.dev/gemini-api/docs/live-api/live-translate
4. Enable input and output transcription when the product needs captions, logging, or a readable translation alongside audio. — https://ai.google.dev/gemini-api/docs/live-api/live-translate

Best practices¶

• Treat it as a continuous translation stream rather than a turn-based agent. Translation mode has no tool, instruction, text-input, image, or video support. — https://ai.google.dev/gemini-api/docs/live-api/live-translate
• For browser or mobile clients, use v1beta ephemeral tokens and constrain translationConfig server-side unless users must choose the target language themselves. — https://ai.google.dev/gemini-api/docs/live-api/live-translate
• Test non-native accents, rapid language switching, long pauses, multi-speaker calls, and noisy rooms before relying on translated audio operationally. — https://deepmind.google/models/model-cards/gemini-3-5-audio/
• Budget for both streamed input and generated output audio: Google lists an effective paid rate of about $0.0368 per minute. — https://ai.google.dev/gemini-api/docs/pricing

Superseded by this¶

• 2026-06-09 — Google planned to replace the prior Google Meet speech-translation limit of five languages and English-centered translation with 70+ languages and 2,000+ language combinations.

Still unknown¶

• The current LiveKit and Pipecat pages describe generic Gemini Live integrations, not a documented Gemini 3.5 Live Translate-specific adapter or configuration. Validate that framework path before treating either as translation support.
• Google’s public material establishes public preview for developers, but does not provide a general-availability date for the API model.

Sources¶

Agent brief¶

• Subject: project:gemini-3-5-live-translate, thread gemini-live-translation, 1 dated events 2026-06-13 → 2026-06-13.
• Practical note: From 2026-06-13, evaluate Gemini 3.5 Live Translate preview through the Gemini Live API or AI Studio and confirm the LiveKit or Pipecat integration path before adopting it.
• Confidence: high. Dated supersedes above are the authority for what is obsolete.