90-minute webinar live captions: AssemblyAI bills open socket time

A 90-minute webinar with live captions costs $0.675 on AssemblyAI's Universal-3.6 Pro Realtime ($0.45 per hour x 1.5 hours) and $0.225 on Universal-Streaming ($0.15 per hour x 1.5). AssemblyAI says streaming is billed on session duration, meaning the time the WebSocket connection is open, not the duration of audio you send. Sume STT cannot caption live; after the event it would cost $0.90 for 90 minutes.

Prices were read on AssemblyAI's pricing page and Sume's catalog on 2026-10-09.

Why the billing basis matters

Most per-hour prices look alike, but the unit differs. If a vendor bills the open connection, then a socket left open for the 20-minute break before the Q&A is billable even though no one is speaking. If you close the socket during breaks and reopen it, you save that time but must handle reconnection and a transcript that restarts.

AssemblyAI's page states the session rule for its streaming models. Other vendors may bill differently, so read each one's billing note rather than assuming.

Where Sume fits

Sume STT is asynchronous. It takes a public HTTPS audio file of up to 10 minutes of reserved duration per job and returns text plus word timings. It does not take a live stream, so it cannot produce captions while the webinar runs.

It does fit the second step. Record the webinar, then transcribe in ten-minute slices after the event: 90 minutes is nine jobs and 90 x $0.01 = $0.90. The word timings can drive burned-in captions for the replay clip.

A simple plan

Use a streaming vendor for the live captions and Sume for the replay assets you make afterward. Set the socket policy up front: decide whether the break time is worth paying for, and test that your client closes the connection when the event ends, because a forgotten session keeps billing on a session-duration basis.

• Live: price by session minutes, not by speech minutes.
• Replay: transcribe the recording once in ten-minute slices.
• Check both vendors' billing notes again before a large event.

Checking the bill after the event

Whatever vendor you use, compare its invoice line with your own log of when each session opened and closed. For a 90-minute webinar, an invoice that shows more than 1.5 hours of session time means a connection was left open or reconnected without closing the earlier one. On a session-duration basis that is the most common way a small event becomes an unexpectedly large bill.

With Sume's batch approach the check is simpler. Each ten-minute slice is its own job with its own audio length, so you can add up the minutes you submitted and compare them to the charge at $0.01 per minute.