AssemblyAI Universal-Streaming $0.15 vs 3.6 Pro $0.45: 250 hours
AssemblyAI lists Universal-Streaming at $0.15 an hour and Universal-3.6 Pro Realtime at $0.45, a 3x step. Over 250 session hours it is $37.50 against $112.50.

AssemblyAI lists its Universal-Streaming models at $0.15 per hour and Universal-3.6 Pro Realtime at $0.45 per hour, so the Pro model costs three times as much. Over a month with 250 session hours, that is $37.50 against $112.50, a difference of $75. AssemblyAI says streaming is billed on session duration, the time the WebSocket is open, so the 250 hours means connection time, not speech time.
Prices were read on AssemblyAI's page on 2026-10-09. Sume STT is a batch service and is shown only for the after-the-fact case.
The month in numbers
The comparison includes AssemblyAI's streaming add-ons where they apply. Speaker diarization is +$0.12 per hour on streaming. Keyterms prompting is included on U3.6 Pro and Multilingual but +$0.04 per hour on Universal-Streaming English.
| Option | Per hour | 250 hours |
|---|---|---|
| Universal-Streaming (English or Multilingual) | $0.15 | $37.50 |
| Universal-Streaming English plus keyterms | $0.19 | $47.50 |
| Universal-3.6 Pro Realtime | $0.45 | $112.50 |
| Universal-3.6 Pro Realtime plus diarization | $0.57 | $142.50 |
| Sume STT 1.0 on 250 recorded hours (batch) | $0.60 | $150.00 |
Choosing between the two tiers
The extra $0.30 per hour buys whatever the Pro model does better on your audio. The price page does not say how much better, so a trial on your own calls is the only way to know. A quick test: stream the same ten recordings through both and count the words you would have to correct.
Keep in mind that keyterms are bundled into the Pro price but extra on the English Universal-Streaming model. If you were going to use keyterms anyway, the Pro model costs $0.45 against $0.19, and the real gap is $0.26 per hour, or $65 over 250 hours.
Where Sume fits
Sume does not offer live transcription, so it is not a substitute for either streaming row. It can transcribe the recorded sessions afterward in ten-minute jobs at $0.01 a minute. That makes sense only if the transcript is the first step of a longer Sume workflow, such as captioned replay clips. For a transcript alone, the batch prices on AssemblyAI's own page are lower than $0.60 an hour.
- Cap session length in your own code to stop forgotten connections.
- Log open and close times so you can reconcile the invoice.
- Use the cheaper tier for rough live captions and the Pro tier only where accuracy matters.
Planning a month of sessions
Count session hours from your calendar, not from speech. 250 session hours is a bit over ten days of one connection open around the clock, or 2,500 sessions of six minutes. Short sessions are cheap one by one but add up, and a pool of agents that keep sockets open between calls multiplies the hours.
If your usage is spiky, set an alert on spend per day. A bug that reconnects in a loop can open many sessions at once, and on a session-duration basis each of them bills until it closes.
Sources
Related posts
More in Comparisons
- Cheapest 4-second AI clip on Sume: the minimum-length math
A 4-second clip at 480p costs $1.07 on Seedance 2.5, $0.25 on Wan 3.0 and $0.15 on Gemini Omni Flash at 360p. MiniMax H3 starts at 5 seconds.
- Cheapest AI video API per second: eight listed prices, sorted
The lowest per-second prices on fal, Runway, Google, Replicate and Sume, read 2026-10-09, from $0.025 to $0.09, with which ones Sume actually lists.
- Cheapest possible clip per Sume video model: the minimum-length floor
The cheapest legal clip on Sume ranges from $0.11 (Omni, 3 s at 360p) to $1.07 (Seedance 2.5, 4 s at 480p). Per-second rates hide the minimum-duration floor.
- Codex custom Responses provider: can it point at Sume?
Codex CLI 0.162.0 adds custom Responses-compatible provider config. Sume's Agent Completions is not a model endpoint; use Sume's hosted MCP from Codex instead.
Written by Sume