Google Launches Gemini 3.5 Transcribe with 4.0% Word Error Rate Across 85 Languages
Google has released Gemini 3.5 Transcribe, a speech-to-text model that automatically detects 85 languages, removes filler words, and corrects misspoken phrases. The company claims a 4.0 percent word error rate for streaming audio and 70 percent lower latency than its predecessor, Chirp 3.
Google has launched Gemini 3.5 Transcribe, a speech-to-text model designed for real-time and recorded audio transcription across more than 85 languages. The model automatically detects the spoken language, removes filler words such as "um," corrects verbal slips, and formats output text without manual intervention.
According to Google, the model achieves a word error rate of 4.0 percent for streaming audio and 2.6 percent for recorded audio. The company also claims 70 percent lower latency compared to its predecessor, Chirp 3, though Google has not published the specific latency figures in milliseconds or the testing methodology behind these numbers.
Two Interfaces, One Model Family
Gemini 3.5 Transcribe ships in two variants tailored to different use cases:
- gemini-3.5-transcribe-live: Accessed via the Live API, this variant handles real-time streaming transcription with low latency, aimed at applications like live captioning or voice interfaces.
- gemini-3.5-transcribe: Accessed via the Interactions API, this variant processes recorded audio and adds speaker attribution and timestamps, useful for meeting transcripts and podcast processing.
The model also supports function calling, allowing it to hand off tasks — such as triggering image generation or web searches — to other models in the Gemini family mid-transcription. Google has not disclosed pricing per 1M tokens or per audio minute for either interface.
Availability
Gemini 3.5 Transcribe is live now in Google AI Studio and on the Gemini Enterprise Agent Platform. It is already integrated into Gboard for Android through a feature called "Rambler" and into the Gemini app on macOS. Google says Chrome support is coming soon, though no specific date has been provided.
Google has not disclosed the model's parameter count, training data cutoff, or architecture details. The 4.0 percent and 2.6 percent word error rate figures, along with the 70 percent latency improvement claim, come directly from Google and have not been independently verified against third-party benchmarks like those maintained by Hugging Face's Open ASR Leaderboard.
What this means
Speech-to-text has become a proving ground for foundation model companies looking to extend beyond text generation into full multimodal pipelines. By building function calling directly into a transcription model, Google is positioning Gemini 3.5 Transcribe as more than a passive dictation tool — it becomes a routing layer that can trigger other Gemini capabilities based on what a user says out loud.
The 85-language claim and sub-3-percent error rate on recorded audio, if verified independently, would put Gemini 3.5 Transcribe in competitive range with OpenAI's Whisper-based systems and specialized providers like ElevenLabs and AssemblyAI. The immediate integration into Gboard and the macOS Gemini app suggests Google is prioritizing consumer-facing deployment over standalone API adoption first — a strategy that could pressure smaller transcription-focused startups on distribution rather than raw accuracy. The lack of published pricing, however, makes it difficult to assess how this fits into Google's broader API cost structure compared to Chirp 3.
Related Articles
Google DeepMind Launches Gemini 3.5 Transcribe, Claims 2.6% Word Error Rate in Testing
Google DeepMind has released Gemini 3.5 Transcribe, a speech-to-text model available via two APIs for real-time streaming and pre-recorded audio. According to Artificial Analysis benchmarks cited by Google, the model achieves a 2.6% word error rate for non-streaming transcription and 4.0% for streaming.
Google Launches Gemini 3.5 Transcribe, a Speech-to-Text Model That Cleans Up Rambling Speech
Google has released Gemini 3.5 Transcribe, a new speech-to-text model that automatically detects over 85 languages, removes filler words, and structures unstructured speech into clean text. The model powers Android's Rambler feature and is rolling out to Chrome, Docs, Gmail, and other Google products.
Google Launches Gemini 3.5 Transcribe with 2.6% Word Error Rate, Powers Gboard Rambler
Google has released Gemini 3.5 Transcribe, a speech-to-text model claiming a 4.0% word error rate in streaming mode and 2.6% in non-streaming mode, according to benchmarks from Artificial Analysis. The model already powers Gboard Rambler on Android and the Gemini app for macOS, with Chrome support coming next.
Z.ai's GLM-5.3-Flash Matches Top Models at 7.5x Lower Cost, Runs Entirely on Chinese Chips
Z.ai released GLM-5.3-Flash, a 320-billion-parameter MoE model with an 18-billion active parameter count and a one-million-token context window. It nearly matches the larger GLM-5.3 on Artificial Analysis's Intelligence Index while costing roughly 7.5 times less per task, and it reportedly runs entirely on Chinese AI chips instead of Nvidia GPUs.
Comments
Loading...