Meta has released its first tool for converting speech to text as it happens, called Muse Voice Transcribe. It comes from Meta Superintelligence Labs, the research group Meta built after paying more than 14 billion dollars for a stake in Scale AI and hiring its founder, Alexandr Wang, to run Meta's AI efforts.
The tool does three things at once that usually require separate software. It writes down what is being said as it is said, it figures out which of more than 20 people in a room is talking at any moment, and it knows when someone has actually finished a sentence rather than just pausing to think. It also handles conversations that mix languages mid sentence, a common pattern for multilingual teams and customers.
The price is what makes this worth paying attention to. Meta is charging three dollars for every 1,000 minutes of audio, which comes out to about 18 cents an hour. That undercuts typical cloud transcription pricing by a wide margin, and it puts real pressure on the meeting-notes and call-transcription tools that companies currently pay for, such as Otter, Fireflies, and Gong.
On accuracy, Meta is not just competing on price. An independent benchmark from Artificial Analysis measured Meta's error rate at 3.1 percent, ahead of Google's Gemini transcription model at 4 percent, and ahead of OpenAI and ElevenLabs as well. For any business that relies on getting spoken words right, whether that is a claims call, a multilingual sales meeting, or a customer service line, that gap in accuracy is the difference between a usable transcript and one that needs a person to fix it afterward.
Where this tool actually shows up matters too. Right now it powers dictation inside Meta's Mac app, where holding one key lets you talk into any application, and it powers Meta's coding assistant. Beyond that, it is available through Meta's developer API, meaning any company can build it into their own call center software, meeting tool, or customer service system. Google, by contrast, is building its competing model directly into Android phones and eventually Chrome, which puts it in front of far more people without anyone having to build anything.
That difference in distribution is worth watching. Meta has the better benchmark numbers and the lower price, but Google has the bigger built-in audience. For a business owner, the practical takeaway is that the cost of adding accurate, multilingual, real-time transcription to any product or workflow just dropped sharply, whether that ends up running on Meta's model, Google's, or whichever one wins the argument on the small print of accuracy and price.
This also tells us something about Meta Superintelligence Labs itself. The division was set up with talk of building superintelligence, but its actual output so far has been narrower and more practical: a coding assistant, an open model, a Mac app, and now a transcription tool. That is not a criticism. Practical tools that businesses can actually use and afford tend to matter more, day to day, than any claim about intelligence.