Meta Superintelligence Labs Muse Voice Transcribe launch

Muse Voice Transcribe, released September 1, 2026, is Meta Superintelligence Labs' first real-time audio perception model. It delivers streaming ASR with a 3.1% word error rate, speaker diarization for more than 20 speakers, and turn-endpointing natively within a single model. The model processes audio in 80ms chunks, dynamically deciding after each chunk whether to emit text or continue listening. It supports multilingual speech and seamless code-switching, with accuracy improvements available through language, keyword, and context biasing. Meta reports it ranks first on Artificial Analysis streaming speech-to-text and public diarization benchmarks. It is available through the Meta Model API, Meta AI for Mac, and Muse Code. On September 4, Meta also released Muse Spark 1.3 with a max reasoning tier on Muse Code and the Meta Model API, reporting stronger coding and agentic performance than the previously available high and xhigh tiers.