Meta unveiled Muse Voice Transcribe on Sept. 1, a real-time AI model that transcribes speech and distinguishes speakers. The company said it can handle conversations involving more than 20 people and recognise multilingual and mixed-language dialogue, including Korean. Developed by Meta Superintelligence Labs, it processes audio in 80-millisecond units and uses adaptive latency to balance accuracy and speed. Meta said it ranked first in an Artificial Analysis streaming test with a 3.1 percent word error rate.

Meta Platforms plans to launch within weeks a consumer version of its conversational AI agent, OpenClaw, which it internally calls Hatch, The Information reported, citing internal documents. The company is also preparing a new AI model, Watermelon, targeting an October release, the report said. Hatch is trained to access websites including DoorDash, Etsy, Reddit, Yelp and Outlook. Meta is considering tiered pricing, with a premium monthly subscription up to $199.99.