Google DeepMind Releases Gemini 3.5 Live Translate for Real-Time Speech Translation Across 70+ Languages
Google DeepMind released Gemini 3.5 Live Translate, an audio model that provides near real-time speech-to-speech translation across 70+ languages. The model automatically detects languages, preserves speaker intonation and pacing, and maintains a few seconds of latency while generating continuous speech output.
Gemini 3.5 Live Translate — Quick Specs
Google DeepMind Releases Gemini 3.5 Live Translate for Real-Time Speech Translation Across 70+ Languages
Google DeepMind released Gemini 3.5 Live Translate on June 9, 2026, an audio model that provides near real-time speech-to-speech translation across 70+ languages with automatic language detection.
Technical Capabilities
The model generates continuous translated speech while maintaining a latency of "just a few seconds" behind the speaker, according to Google. Unlike turn-based translation systems that wait for complete sentences, Gemini 3.5 Live Translate processes streaming audio and balances translation speed with contextual accuracy.
Key technical features include:
- Automatic detection of 70+ languages without manual configuration
- Preservation of speaker intonation, pacing, and pitch in translated output
- Noise robustness for unpredictable environments
- Support for over 2,000 language pair combinations in single sessions
- SynthID watermarking embedded in all generated audio
Availability and Deployment
Gemini 3.5 Live Translate is rolling out across three channels:
Gemini Live API: Available in public preview for developers via Google AI Studio. Developer platforms including Agora, Fishjam, LiveKit, Pipecat, and Vision Agents have integrated the API for real-time media streaming infrastructure.
Google Meet: Launching in private preview this month for select Google Workspace business customers, expanding from the previous limitation of five languages and English-only translation pairs. Broader rollout planned for later in 2026.
Google Translate app: Rolling out globally on Android and iOS. The model powers the Live translate feature for users with connected headphones. Android users receive an additional "listening mode" that streams translations through the phone's earpiece without headphones.
Early Implementations
Grab, which processes over 10 million voice calls monthly, is testing the model to enable multilingual communication between drivers and travelers. Additional partners including CJ ENM and LiveKit have provided feedback on translation quality and low latency, according to Google.
Pricing for API access has not been disclosed.
What This Means
Gemini 3.5 Live Translate represents Google's entry into the competitive real-time speech translation market, directly challenging established players in multilingual communication tools. The 70+ language support and 2,000+ language pair combinations significantly exceed the capabilities of Google's previous Meet translation system, which supported only five languages with English as a required pivot.
The model's continuous streaming approach addresses a core limitation of turn-based systems, though the "few seconds" latency specification lacks precision for developers evaluating real-time requirements. The integration across Google's product ecosystem—from developer APIs to consumer apps—indicates a platform play rather than a standalone model release. However, the lack of disclosed API pricing and benchmark comparisons to competing speech translation models limits technical evaluation.
Related Articles
Chariot launches Chariot-1-ASR Realtime, a streaming speech recognition model for 10 Indian languages
Chariot, a frontier AI research lab, launched Chariot-1-ASR Realtime on October 22. The streaming speech recognition model covers 10 Indian languages plus Indian English. The company claims it made fewer errors than every streaming and batch system it tested, based on its own evaluations.
MiniMax releases M3.1-Flash: 428B-parameter multimodal MoE with 1M-token context and weights on Hugging Face
MiniMax has published MiniMax-M3.1-Flash on Hugging Face: a native multimodal mixture-of-experts model with about 428B total parameters, 23B activated per token, and a context window of up to 1M tokens. It builds on MiniMax-M3 with sparse attention, quantization-aware training and DSpark speculative decoding.
Tencent releases WeVisOne-VL-4B, a Qwen3-VL-based model for object detection, segmentation and alpha matting
Tencent's WeChat Vision team released WeVisOne-VL-4B, a vision-language model built on Qwen3-VL that handles object-level and pixel-level perception. It adds a 9.6M-parameter pixel decoder to produce segmentation masks and full-resolution alpha mattes.
Tencent releases WeVisOne-VL-2B, a 2B-parameter model for object detection, segmentation and alpha matting
Tencent's WeChat Vision team released WeVisOne-VL-2B, a 2B-parameter vision-language model built on Qwen3-VL. It handles object-level and pixel-level perception, including referring segmentation and alpha matting, through structured next-token prediction and a 9.5M-parameter pixel decoder.
Comments
Loading...