Google has introduced Gemini 3.5 Transcribe — and it’s the company’s most precise speech-to-text model yet. Specifically, the model is built for intelligent voice interactions. Rather than relying on older methods, it turns raw audio directly into accurate, polished, formatted text.
Takeaways 🚀
✅Precise real-time transcription by Google
✅Handles noise, jargon, disfluencies
✅Supports multilingual speech recognition
✅Adds speaker labels, timestamps
✅Improves voice app workflows
Traditional speech recognition tools often struggle in real-world conditions. For instance, background noise, complex jargon, and messy speech can easily trip them up. Gemini 3.5 Transcribe, however, handles all of these challenges with ease.
Why This Model Stands Out
People are already using this technology across Google products. For example, you’ll find it in the Gemini app and on Android. New features like Rambler on Android demonstrate what it can do, and the Gemini app on macOS uses it as well.
Beyond consumer products, developers can now build their own tools with it. The model is available through the Gemini API in Google AI Studio and also works with the Gemini Enterprise Agent Platform.
Two APIs for Different Needs
Gemini 3.5 Transcribe fits seamlessly into your existing workflows. Whether you’re building voice agents, live captioning tools, or post-call analytics pipelines, the model supports you through two separate APIs.
Real-time streaming. This option delivers continuous, two-way streaming with sub-second latency for interactive voice apps. Developers can access it via the Live API using gemini-3.5-transcribe-live.
Pre-recorded audio processing. This option, on the other hand, transcribes recorded audio, meetings, and call logs. In addition, it adds speaker attribution and word-level timestamps. Developers use the Interactions API with gemini-3.5-transcribe.
Precise and Intelligent Transcription
At its core, the model captures your natural speaking style and understands your intent. It also recognizes custom vocabulary, so you can complete tasks with your voice more efficiently.
Here’s what makes it especially smart:
- Smart transcription. The model handles self-corrections effortlessly. For example, say “let’s meet Tuesday—no, Wednesday,” and it gets it right. Furthermore, it removes filler words like “ums” and “ahs” and auto-formats your text.
- Function calling. In addition to transcribing, the model can pass complex tasks to other Gemini models — such as image generation and file analysis. This feature currently works in the Gemini macOS app.
- More precise transcription. According to Artificial Analysis, the model achieves an average Word Error Rate of 4.0% for streaming and 2.6% for non-streaming use cases. Moreover, it performs well in noisy, real-world settings and accurately captures alphanumeric details like postal codes and order IDs.
- Custom vocabulary. The model recognizes specialized jargon and unique spellings, and it adapts to the custom vocabulary you provide.
- Global language support. The model automatically detects and transcribes over 85 languages. On top of that, it handles regional accents and diverse dialects with ease.
- Multi-speaker identification. Finally, the model attributes speech in pre-recorded audio with timestamps for up to three speakers. Support for more than three speakers is still experimental.
A Big Leap From Chirp 3
Compared to its predecessor, Gemini 3.5 Transcribe marks a major step forward. It builds on Chirp 3, the previous transcription model, by offering new capabilities, better word error rates, and much faster performance overall.
According to Artificial Analysis, the time to final transcription improved by 70%. Additionally, the model performs strongly on the FLEURS benchmark. Across top languages and locales, it reaches a 5.50% WER in streaming mode and 5.04% WER in non-streaming use cases.
Smart Transcription Across Google
Beyond speech-to-text, the model brings context-aware understanding to everyday tools. As a result, you’ll find it in Gboard, Antigravity, the Gemini app, and Chrome — where it captures nuance, intent, and inline edits with ease.
Gboard on Android. Through the new Rambler feature, the model turns spoken thoughts into clean, formatted text and filters out filler words. In addition, you can use your voice to make edits, fix misspellings, and change the writing style.
Google Antigravity. Here, the model pairs screen context with chat history, with your permission. This ensures pinpoint accuracy across file names, agent thoughts, and active documents.
Google AI Studio. In this environment, you can access the model in Build mode, which lets you code apps with your voice on the fly.
Gemini app on macOS. Not only does the model transcribe your natural speech into clean text, but it also enables voice commands that work alongside screen context. It calls on other Gemini models in the background to handle the heavy lifting — so you can summarize local files, repurpose text across apps, or generate images at your cursor, all with your voice.
Chrome (coming soon). Soon, you’ll be able to talk to type in any web field. As a result, dictating replies, drafting posts, or prompting Gemini in Chrome will feel more natural than ever.
Early Reviews From Developers
Several developer platforms are already using the Gemini Live API. These include Agora, Fishjam, LangChain, LiveKit, Pipecat, Vercel, and Vision Agents. Together, they let developers build high-performance voice interfaces with ease, since these platforms manage complex real-time media streaming behind the scenes. Consequently, developers can focus entirely on crafting the user experience.
In addition, companies like Vivo, Intellitek Health, and Lingopal have shared positive feedback. Specifically, they praised the model’s low latency, strong accuracy, and wide language support.
Start Using Gemini 3.5 Transcribe
The model is ready for you to try today. Here’s how to get started, depending on who you are.
For developers. You can access it in public preview through the Gemini API via Google AI Studio and Google Antigravity.
For enterprises. It’s also in public preview via the Gemini Enterprise Agent Platform. Support for Gemini Enterprise for Customer Experience is coming soon.
For everyone. You can use it in the Gemini app on macOS in English. Meanwhile, Rambler on Android is available in select countries and languages, and Chrome support is on its way.
Ultimately, Gemini 3.5 Transcribe brings speech-to-text a step closer to natural conversation. Whether you build apps or simply talk to your devices, the model makes voice feel effortle
