Meta launches Muse Voice Transcribe, a real-time speech recognition model

Archive entry: this covers an event from September 1, 2026, and we published it on November 8, 2026. The details come from sources dated at the time and may have changed since.

On September 1, 2026, Meta Superintelligence Labs released Muse Voice Transcribe, its first real-time audio perception model, according to the Meta AI Research announcement. The model offered streaming speech recognition, diarization for 20+ speakers and endpointing, and it was priced at $0.18 per hour of audio.

What changed

• It supported code-switching and language, keyword and context biasing.
• It was trained on 70+ languages, 25 of which were verified.
• It was available through the Meta Model API, Meta AI for Mac and Muse Code.
• The model ID was muse-voice-transcribe-1.0, with the endpoints wss://api.meta.ai/v1/asr/realtime and POST /v1/asr/transcribe.
• Pricing was $0.18 per hour of audio, with limits of 256 concurrent streams and 16,000 streams per hour.

What it means for people building with AI

Meta entered paid speech-to-text APIs with this release, at $0.18 per hour of audio. Meta said the model ranked #1 on Artificial Analysis for streaming speech-to-text and diarization as of September 1, 2026. Builders also needed to account for one gap: the model did not provide word-level timestamps.

Documentation

Meta's developer documentation for the Meta Model API includes a models page and a pricing and rate limits page.

Sources

• Introducing Muse Voice Transcribe (Meta AI Research, September 1, 2026, official)
• Models (Meta (dev.meta.ai docs), official)
• Pricing and rate limits (Meta (dev.meta.ai docs), official)

Written with AI assistance, strictly from the sources listed above. Before it went live, automated checks verified its dates, sources and every language edition.