Google has introduced Gemini 3.5 Transcribe, a speech-to-text model for live conversations, recorded meetings and voice-driven software. The company says the model produces formatted text directly from raw audio and is built to handle noise, specialist vocabulary, accents and spoken corrections.
The release comes in two versions for developers. The Live API model, called gemini-3.5-transcribe-live, supports two-way audio streaming with sub-second latency for interactive applications. A separate Interactions API model, gemini-3.5-transcribe, processes recordings and can add word-level timestamps and speaker attribution.
Google says the model can remove filler words, interpret revisions made while speaking and format text automatically. For example, it can reflect a speaker changing “Tuesday” to “Wednesday” without retaining the abandoned instruction. It also supports custom vocabulary, which could help teams transcribe company names, technical terms, product codes and other uncommon words.
Accuracy, languages and speakers
According to Google, Gemini 3.5 Transcribe detects and transcribes more than 85 languages. For recorded audio, it identifies up to three speakers, while recognition of more speakers remains experimental.
The company cites testing by Artificial Analysis that gives the model an average word error rate of 4.0 percent for streaming transcription and 2.6 percent for non-streaming work. Google also says the time to final transcription is 70 percent lower than with its earlier Chirp 3 model. These are company-reported comparisons and benchmark results.
Google has already deployed the technology in its Gemini app for macOS and in Rambler, a Gboard feature on Android. In the macOS app, it can also trigger function calls to other Gemini models for tasks such as analysing files or generating images. Google says voice typing powered by the model will also arrive in Chrome.
The developer and enterprise versions are in public preview through Google AI Studio, Google Antigravity and the Gemini Enterprise Agent Platform.
Sources
- Intelligent transcription with Gemini 3.5 Transcribe – Google
- Google debuts Gemini 3.5 Transcribe, a speech-to-text model that powers Gboard Rambler and is coming to Chrome, in public preview for developers and enterprises – 9to5Google
Stay up to date
AI for content creation: the latest tools, tips and trends. Every two weeks in your inbox:
