Google Gemini 3.5 Transcribe is the company’s newest AI model for speech-to-text, built to convert spoken words into clean, polished text by trimming filler sounds and on-the-fly corrections. While the anticipated Gemini 3.5 Pro remains unreleased, Google is rolling out this transcription-focused model across its ecosystem, starting with the Gboard “Rambler” feature on the Pixel 11.
According to Google, the new model is significantly faster and more accurate than its predecessor, the Chirp 3 engine. Gemini 3.5 Transcribe delivers voice-to-final-text results roughly 70 percent faster, and its live-speech error rate has dropped to 5.5 percent. That marks an improvement over Chirp 3, which Google measures at 7.32 percent. For anyone who relies on voice input, even a modest gain in accuracy reduces the tedium of fixing typos after dictation.
What Gemini 3.5 Transcribe Can Do
Beyond capturing words more accurately, the model aims to interpret intent. As you speak, it can remove awkward “ums” and “uhs” that clutter natural speech, edit text in real time when you correct yourself, and reference custom vocabulary to handle specialized jargon. The feature supports 85 languages and can distinguish up to three speakers in pre-recorded audio.
The tradeoff is that users depend on the AI to grasp the meaning of their speech. For short passages, the model handles inconsistencies and verbal stumbles well, but it does technically alter the wording of what was said, which may not suit every situation.
Where the Model Is Available
Gemini 3.5 Transcribe is already live in several places, including Rambler in Gboard, though that remains limited to Pixel 11 phones for now. Google says it will expand Rambler to more Gemini Intelligence devices later this year. Users of the Gemini app on macOS also gain the upgraded voice input starting today.
Developers get broad access as well. Antigravity now includes the new model, with full access to screen context and chat history when granted permission. AI Studio’s build model has adopted Gemini 3.5 Transcribe, enabling voice-driven app development with AI-optimized transcription. The model is also available through the Gemini API.
For those not using these specific apps or devices, the feature is still on the way. Google plans to bring AI transcription to the Chrome browser soon, allowing text input into any web field for tasks such as writing emails, leaving comments, or prompting AI chatbots by voice.
Source
Image: arstechnica.com
