--- title: "Google launches Gemini 3.5 Transcribe with sub-second streaming and 85+ language support" url: https://daily.dev/posts/google-launches-gemini-3-5-transcribe-with-sub-second-streaming-and-85-language-support-cef0b9qnr source_url: https://daily.dev/posts/google-launches-gemini-3-5-transcribe-with-sub-second-streaming-and-85-language-support-cef0b9qnr type: collection source: "Collections" published: 2026-08-26T17:21:58.109Z updated: 2026-08-26T21:57:48.804Z tags: ["google", "google-gemini", "speech-recognition"] reading_time: 2 upvotes: 0 comments: 0 language: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # Google launches Gemini 3.5 Transcribe with sub-second streaming and 85+ language support **[Collections](https://daily.dev/sources/collections)** · 2 min read · 0 upvotes · 0 comments ## Summary Google has released Gemini 3.5 Transcribe, a new speech-to-text model in public preview via the Gemini API and Google AI Studio, with two variants: a streaming model (gemini-3.5-transcribe-live) offering sub-second bidirectional transcription, and a batch model (gemini-3.5-transcribe) with word-level timestamps and speaker attribution for up to three speakers. Compared to Chirp 3, it achieves 2.6% word error rate on non-streaming audio, 4.0% on streaming, a 70% cut in time-to-final-transcription, and auto-detects 85+ languages. Notable post-processing features include filler-word removal, self-correction handling, and contextual alphanumeric interpretation (e.g., recognizing spoken ## Content Google has released Gemini 3.5 Transcribe, its first Gemini-based speech-to-text model, now in public preview via Google AI Studio and the Gemini API. ## What it is Unlike traditional transcription models, Gemini 3.5 Transcribe is LLM-based, which means it handles the messy parts of real speech better than its predecessors. It strips filler words like "um" and "ah", resolves mid-sentence self-corrections, and formats things like phone numbers, postal codes, email addresses, and unit conversions correctly. One engineer noted it knows you mean `.json` and not a person named "Jason" - that kind of contextual awareness is what separates it from older approaches. It supports 85+ languages with automatic detection, and can identify up to three speakers with word-level timestamps on pre-recorded audio. ## Two endpoints There are two ways to use it: - **gemini-3.5-transcribe** - for recorded audio up to one hour, with timestamps, speaker attribution, and custom vocabulary support, available via the Interactions API - **gemini-3.5-transcribe-live** - real-time bidirectional streaming via the Gemini Live API, with sub-second latency, capped at 10 minutes per session ## How it compares to Chirp 3 On benchmarks, it hits 2.6% word error rate in non-streaming mode and 4.0% in streaming, outperforming Chirp 3 on FLEURS. Time-to-final-transcription is 70% faster than Chirp 3. It also auto-detects over 70 languages even when only English hints are set, and supports custom vocabulary to improve accuracy for names and domain-specific terms. ## Where it's already running The model is already powering Gboard's Rambler feature and the Gemini app on macOS. It's coming to Chrome. Third-party platforms including LiveKit, Pipecat, and Vercel have already integrated it - Vercel's AI Gateway supports it now with automatic language detection across 85+ languages. ## Building with it For developers, Google has published a walkthrough for building a live speech translation app using the Gemini Live API combined with LiveKit for WebRTC streaming and Cloud Run for deployment. The open-source Next.js demo spins up one translation session per target language and reuses sessions for additional listeners rather than duplicating them. One practical note: the demo runs on a single Cloud Run instance because translation state lives in memory, which limits it to roughly 15-20 simultaneous languages before hitting CPU limits. Moving state to a database would allow multi-instance scaling. Access is available now through Google AI Studio and the Gemini API. ## Questions this post answers ### What is the word error rate of Google's Gemini 3.5 Transcribe compared to Chirp 3? Gemini 3.5 Transcribe achieves a 2.6% word error rate on non-streaming audio and 4.0% on streaming audio, compared to Chirp 3, while also cutting time-to-final-transcription by 70%. It auto-detects more than 85 languages and comes in two variants: gemini-3.5-transcribe-live for sub-second bidirectional streaming, and gemini-3.5-transcribe for batch processing with speaker attribution. _Developers picking a speech-to-text API can track model benchmarks like these on daily.dev._ ### Does Gemini 3.5 Transcribe support speaker identification and timestamps? Yes, the batch variant gemini-3.5-transcribe provides word-level timestamps and speaker attribution for up to three speakers, accessed via the Interactions API. The streaming variant, gemini-3.5-transcribe-live, instead focuses on sub-second bidirectional transcription through the Gemini Live API without these batch-only features. _Teams evaluating transcription APIs for their stack can follow these capability details on daily.dev._ ### Where is Google's Gemini 3.5 Transcribe available besides the API? Beyond the Gemini API and Google AI Studio, it is already live in the Gemini app on macOS, Google Antigravity, Gboard's Rambler feature, and the Gemini Enterprise Agent Platform, with Chrome support coming. Third-party platforms LiveKit, Pipecat, and Vercel are also integrating the model. _Developers wiring voice features into their products can track rollout details like these on daily.dev._ ## Community take How the wider developer community reacted, aggregated from 5 discussions and 8 comments across x (as of 2026-08-26). **TL;DR:** Reaction is light and mostly casual enthusiasm, with a few asking what practical improvement it offers over existing tools and whether the real goal is voice-native agents rather than plain transcription. **Sentiment:** 45% positive · 40% mixed · 15% skeptical **The case for** - Combining smart transcription, function calling, and broad language support in one release is seen as a strong package for voice-first apps. - Some found the 'smart mode' impressively capable. **The pushback** - One commenter couldn't identify what the 'significant' improvement actually is compared to other tools. - A request for faster performance implies current speed isn't fully satisfying to everyone. **By community** - x (mixed): Replies are a mix of light hype and a few pointed questions about real-world differentiation and underlying purpose. **Open questions** - Is the underlying goal better transcription accuracy or enabling fully voice-native agents? - What concretely makes this a 'significant' improvement over other transcription tools? - Will a similarly upgraded Gemini 3.5 Pro follow? **Highlights** > @osanseviero No logro entender cual es la mejora "significativa" en comparación de otras herramientas. Me ayudas? Gracias > — [GastonG\_ar on x](https://x.com/GastonG_ar/status/2092720449180373364) > @osanseviero is the real goal transcription or enabling fully voice native agents ? > — [sonilapt on x · 1 points](https://x.com/sonilapt/status/2092672646609854630) > @osanseviero Smart transcription + function calling + wide language support in one release is a strong package for voice-first applications. > — [AIHumanTeam on x](https://x.com/AIHumanTeam/status/2092669944152641850) > @osanseviero Can we make it faster > — [mariusfanu on x](https://x.com/mariusfanu/status/2092708084397301823) **Source threads** - [x](https://x.com/gudanglifehack/status/2092675010494341275) · 1 points · 0 comments - [x](https://x.com/techsnif/status/2092697153906897003) · 0 points · 0 comments - [x](https://x.com/osanseviero/status/2092721170776793182) · 0 points · 0 comments - [x](https://x.com/osanseviero/status/2092667118017740879) · 0 points · 8 comments - [x](https://x.com/cfaryanoconnell/status/2092663128748069218) · 0 points · 0 comments --- Tags: [#google](https://daily.dev/tags/google), [#google-gemini](https://daily.dev/tags/google-gemini), [#speech-recognition](https://daily.dev/tags/speech-recognition) [View this post on daily.dev](https://daily.dev/posts/google-launches-gemini-3-5-transcribe-with-sub-second-streaming-and-85-language-support-cef0b9qnr)