Google Releases Rambler for Free, One of the Coolest Pixel Features
Google has unveiled Gemini 3.5 Transcribe, the model powering the new Rambler feature. It excels at filtering out background noise and filler words, offering significantly improved transcription accuracy.
Two weeks ago, Google unveiled the new Pixel 11 series of devices, and among the various innovations, it revealed a feature called Rambler. Now, the company is officially releasing the model that underlies this feature, so that other devices can enjoy it. Meet Gemini 3.5 Transcribe, Google's new transcription model that brings a real breakthrough.
Its most central ability is to filter out all human background noises. Even if you record a message that consists mostly of "ummm" and pauses, the final output will include only the relevant words, clean of unnecessary filler words. This ability is exactly what puts it at the heart of the Pixel 11's Rambler feature. The name is derived from the English verb meaning "to ramble," and it is intended to save exactly that: to cut out all the unnecessary words around your main message.
The feature, which was first presented at Google I/O in May, has now landed on the company's flagship devices as part of the Gboard keyboard, but at least in the first stage, it is not available to other manufacturers or older Pixel models. In terms of performance, Google reports that the new transcription model is significantly faster and more accurate than its previous engine, Chirp 3. The transcription speed recorded a jump of 70% compared to the previous generation, and the error rate stands at 5.5% — an improvement compared to 7.32% in Chirp 3.
In addition to performance, the model supports more than 85 languages built-in from launch — including Hebrew. It also includes a dedicated window where you can enter specific words that the model might miss, allowing it to learn new terms and save you repeated corrections. The model demonstrates an impressive ability to switch between languages in real time and knows how to identify references to numbers and dates, converting them into a logical numerical format.
Another cool feature is smart transcription. If you dictate: "I want a large espresso, uh, actually, I prefer an americano, no, no, I want a latte with a little milk" — it will offer you both the full transcription and a version of "I want a latte with a little milk."
As of now, the model reaches users through a macOS application and for Pixel 11 users through the Rambler feature. Additionally, the model is available to developers through the API and the AI Studio environment.
Despite the new feature, one cannot ignore Google's delay in major launches. At its last developer conference, it unveiled the Gemini 3.5 family and promised that a month later Gemini 3.5 Pro would arrive. In practice, this deadline passed two months ago, and the latest Pro model remains Gemini 3.1 Pro, launched half a year ago — an eternity in terms of the model life cycle today. While Google continues to release fast Flash models, they cannot truly compete with the advanced models from OpenAI and Anthropic. Furthermore, Chinese players like Z.ai and Alibaba are releasing excellent models with open weights that have already managed to overtake Google.
