Speech-to-text cost per hour: Scribe v2, MAI-Transcribe-2 and Sume

Per hour of audio the list prices are $0.22 for ElevenLabs Scribe v2, $0.39 for Scribe v2 Realtime, $0.54 for Microsoft's MAI-Transcribe-2-Streaming through the end of the year, and $0.60 for Sume's speech-to-text at $0.01 per audio minute.

Hourly table

What else differs

MAI-Transcribe-2-Streaming is a streaming model: Microsoft says first partials arrive in just over 100 ms and that it covers 60 languages. Unite.ai reports 2.5% final word error rate. Those are vendor-reported figures; Sume publishes none of its own here.

Sume's STT is a job, not a stream: it caps at 10 minutes (600 seconds) per job and takes a public https audio_url.

• For a 45-minute recording, split into five jobs of nine minutes or fewer.
• Diarization and audio-event tags are fixed on the server, not request options.
• Sentence segmentation has a boundary_lead_ms of 0 to 500, default 70.

Worked example

A 3-hour archive is 3 x $0.22 = $0.66 on Scribe v2, 3 x $0.54 = $1.62 on MAI-Transcribe-2-Streaming and 180 x $0.01 = $1.80 on Sume STT.

Choosing

Cost per hour is the smallest difference here. Pick on streaming or batch, languages and what you do with the transcript.