TL;DR: Google introduced Gemini 3.5 Transcribe on August 26, 2026, a speech-to-text model that removes filler words, resolves self-corrections and formats dictated speech.
Google introduces Gemini 3.5 Transcribe
Google introduced Gemini 3.5 Transcribe on August 26, 2026, describing it as its most precise speech-to-text model yet for intelligent voice interactions, according to the official Google blog post titled Intelligent transcription with Gemini 3.5 Transcribe. The post was authored by Diego Melendo Casado, Senior Director of Engineering for Gemini Audio, and Luke Leonhard, Chief of Staff for Gemini Audio, on behalf of the Gemini Audio Team. The model converts raw audio directly into accurate, polished, formatted text, unlike conventional speech recognition models that struggle with background noise, complex jargon and disfluency cleanup. Google said consumers were already benefiting from the transcription model through new voice capabilities such as Rambler on Android and in the Gemini app on macOS. Developers can build similar capabilities with Gemini 3.5 Transcribe in the Gemini API in Google AI Studio and the Gemini Enterprise Agent Platform. The Verge reported the same August 26 launch, noting that the addition to the Gemini family follows the launch of 3.5 Live Translate.
The announcement arrived while Google had not yet released the Gemini 3.5 Pro model that it had promised to roll out in June, according to The Verge. The Verge reported that Google describes 3.5 Transcribe as representing a major advancement from the previous transcription model, Chirp 3, especially regarding multilingual performance and wording error rates. Google's own announcement said performance over Chirp 3 includes new capabilities, improved word error rates and significantly better latency. The new model is designed to capture natural speaking style to better understand intent and recognize custom vocabulary so users can execute tasks with voice. Google built 3.5 Transcribe to plug into developer workflows for voice agents, real-time captioning tools and post-call analytics pipelines.
Two APIs cover live and recorded audio
Real-time streaming for interactive voice
Real-time streaming delivers continuous, bidirectional streaming with sub-second latency for interactive voice apps through the Live API using gemini-3.5-transcribe-live, according to Google. The streaming path is intended for interactive voice apps that require continuous transcription as audio arrives. Google said developer platforms such as Agora, Fishjam, LangChain, LiveKit, Pipecat and Vercel were leveraging the Gemini Live API. 9to5Google reported that the model is available for developers in public preview in the Gemini API via Google AI Studio and Google Antigravity. The Verge also reported that the model is available for developers in public preview in the Gemini API via AI Studio and Antigravity. Google also offers enterprise access in public preview via the Gemini Enterprise Agent Platform, according to 9to5Google.
Pre-recorded processing with speakers and timestamps
Pre-recorded audio processing transcribes recorded audio, meetings and call logs with speaker attribution and word-level timestamps through the Interactions API using gemini-3.5-transcribe, according to Google. The pre-recorded path attributes speech in pre-recorded audio with timestamps for up to three speakers, with support for more than three speakers described as experimental. The Verge reported that the model can attribute speech for up to three speakers in pre-recorded audio alongside providing word-level timestamps. 9to5Google reported the same three-speaker attribution and word-level timestamps for pre-recorded material. Google said the model accurately captures alphanumeric entities like postal codes and order IDs and shows strong performance across noisy, real-world environments.
Smart transcription cleans filler words and corrections
Smart transcription seamlessly handles self-corrections such as "let's meet Tuesday — no, Wednesday," removes filler words described as "ums" and "ahs," and auto-formats text, according to Google. The Verge reported that the transcription model allows users to edit naturally with just voice and can automatically format text and remove filler words like "um" and "uh." 9to5Google reported that the model can handle the same Tuesday-to-Wednesday self-correction example and remove ums, ahs and other filler words from the end result while auto-formatting text and allowing natural voice editing. On Gboard on Android, Google said the new Rambler feature transforms spoken thoughts into well-formatted text while filtering out filler words. Users can also use voice to make edits, correct misspellings and change writing style in Rambler, according to Google.
Function calling lets the model delegate complex tasks such as image generation and file analysis to other Gemini models via function calls, according to Google. That function-calling capability is currently available in the Gemini macOS app. In the Gemini app on macOS, Google said 3.5 Transcribe not only transcribes free natural speech into clean formatted text but also enables voice commands that pair with screen context to power complex workflows. By calling on other Gemini models in the background to handle heavier processing, the model can summarize local files, repurpose text across apps and generate images at the cursor using voice, according to Google. 9to5Google described the same macOS Speak to Window capability as an example of executing tasks with voice through function calling.
Custom vocabulary and 85-language support
Users can provide customized vocabulary to the model, allowing 3.5 Transcribe to automatically adapt transcription to unique spelling requirements and specialized jargon to prevent manual editing of those words, according to The Verge. Google said custom vocabulary recognizes specialized jargon and unique spellings by adapting transcriptions to the provided custom vocabulary. 9to5Google reported the same ability to recognize specialized jargon and unique spellings through supplied custom vocabulary. Global language support automatically detects and transcribes over 85 languages while handling regional accents and diverse dialects, according to Google. The Verge reported that the updated Gemini Audio transcription capabilities automatically detect specialized jargon and more than 85 languages. 9to5Google reported the same automatic detection and transcription of over 85 languages with handling of regional accents and dialects.
Google said 3.5 Transcribe handles live language switches and seamless streaming transcription. In Google Antigravity, Google said 3.5 Transcribe pairs screen context and chat history, with permission, to ensure transcription accuracy across file names, agent thoughts and active documents. 9to5Google reported that the Antigravity prompt-box microphone uses the same pairing of screen context and chat history, with permission, for file names, agent thoughts and active documents. In Google AI Studio, users can access 3.5 Transcribe in Build mode to build apps with voice, according to Google. 9to5Google described that AI Studio Build-mode use as vibe-coding apps with voice on the fly.
Accuracy and latency measurements
As measured by Artificial Analysis, Gemini 3.5 Transcribe achieves an average Word Error Rate of 4.0% for streaming and 2.6% for non-streaming use cases, according to Google. 9to5Google reported the same 4.0% streaming and 2.6% non-streaming figures from Artificial Analysis. Google said time to final transcription improves by 70% over Chirp 3, as measured by Artificial Analysis. 9to5Google reported the same 70% improvement in time to final transcription over the Chirp 3 transcription model from 2025. On the FLEURS benchmark across a set of top languages and locales, Google said the model delivers precise multilingual performance, improving over Chirp 3 and achieving a 5.50% Word Error Rate in streaming mode and a 5.04% Word Error Rate in non-streaming use cases. 9to5Google reported the same FLEURS results of 5.50% streaming and 5.04% non-streaming.
Where the model is available now
Gemini 3.5 Transcribe is rolling out starting August 26 in English for all macOS Gemini app users, and through the Rambler dictation feature on Android in select countries and languages, according to The Verge. A separate Google post on August 25, 2026, by Suresh Ganapathy said intelligent dictation in the Gemini app for macOS allows users to speak naturally into any window on the desktop, transcribing spoken words into clean, polished text. That macOS feature automatically removes ums and ahs, catches mid-sentence corrections and drops formatted text at the cursor, according to Google. The August 26 Google announcement said the macOS dictation work was already available to consumers and that developers could access the model in the Gemini API.
Chrome support is planned as a future release, not a current launch, with Google saying users will be able to talk to type in any web field. Google described the planned Chrome use as dictating replies, drafting posts and prompting Gemini in Chrome more naturally with voice. 9to5Google reported that the Chrome browser capability is coming next for talking to type in any web field. The Verge also reported that Google says Chrome support is coming soon. For enterprises, 9to5Google reported that Gemini 3.5 Transcribe is in public preview via the Gemini Enterprise Agent Platform and is coming soon to Gemini Enterprise for Customer Experience, another future availability item.
Live models were not part of the launch
Alongside 3.5 Transcribe, Google had initially indicated that 3.5 Live and 3.5 Live Experimental updates would be coming to Gemini Audio on August 26, building on existing speech recognition technology powering Gemini's voice chat mode, according to The Verge. After publication of that report, Google told The Verge that those additional models are not being launched yet and did not provide a new launch date. According to information Google had previously provided to The Verge, Gemini 3.5 Live is better at handling mid-sentence interruptions, language recognition and live visual processing. The same prior information said Gemini 3.5 Live Experimental goes further by narrating its progress step by step in real time while tackling reasoning on more complex tasks. The Verge added an August 26 update stating that Google mentioned the two Live models in information provided prior to publication but now says only 3.5 Transcribe is being announced.