Deepgram Voice Agent API $0.075 a minute vs AssemblyAI $4.50 an hour
Deepgram's standard Voice Agent tier and AssemblyAI's Voice Agent API are both $0.075 a minute ($4.50 an hour). Deepgram advanced is $0.163. Where Sume fits.

Two all-in voice agent APIs list the same price: Deepgram's standard Voice Agent tier is $0.075 per minute on pay as you go, and AssemblyAI's Voice Agent API is $4.50 an hour, which is also $0.075 a minute. Deepgram's advanced tier is $0.163 a minute. Neither includes anything Sume sells as a live product, because Sume has no live agent; its text-to-speech is about $0.036 to $0.038 per minute of speech and its speech-to-text is $0.01 a minute, both as file jobs.
Numbers are from Deepgram's pricing page and AssemblyAI's pricing page, both read on 2026-10-03. AssemblyAI's page lists what its rate covers; I did not see a comparable description of what Deepgram's tiers include beyond the label, so treat the comparison as price only.
The prices side by side
Per-hour figures are the per-minute price times 60.
| Product | Per minute | Per hour | 10,000 minutes |
|---|---|---|---|
| Deepgram Voice Agent, standard | $0.075 | $4.50 | $750 |
| Deepgram Voice Agent, advanced | $0.163 | $9.78 | $1,630 |
| AssemblyAI Voice Agent API (all-inclusive) | $0.075 | $4.50 | $750 |
What the pages say about the bundle
AssemblyAI describes its price as all-inclusive of STT, LLM reasoning, TTS, turn detection, interruption handling and tool calling. Deepgram lists the two tiers under Voice Agent API on the pay-as-you-go plan, and a separate Growth plan starting at $4K a year with pre-paid credits offering savings of up to 20 percent against pay-as-you-go rates. So the same $750 for 10,000 minutes could be lower on Deepgram at Growth volumes, by up to $150, if the 20 percent applies to this product; the page does not say which products the saving covers.
Setting Sume's lines beside them
Sume text to speech is $0.0475 per 1,000 characters, each quote rounded up to whole cents, up to 20,000 characters or 1,200 seconds of audio per request. Speech to text is $0.01 per minute of audio, up to 600 seconds per job.
Per-minute speech figures assume 750 to 800 characters a minute, the range Cartesia publishes for its voices; a script's real length varies with pace.
A minute of caller audio transcribed plus a minute of agent speech synthesized is $0.01 plus about $0.036 to $0.038, which is $0.046 to $0.048. That sits below the $0.075 bundle, but it leaves out the language model, turn-taking and any phone line, and it is delivered as a file, not a stream.
A support line
Take 2,000 calls a month averaging 4 minutes, 8,000 minutes. Deepgram standard and AssemblyAI both come to $600, and Deepgram advanced to $1,304. Transcribing every call afterwards on Sume STT, each under the 10-minute cap and so one job, is 2,000 x $0.04 = $80. That is a 13 percent addition on the $600 and gives you word-timed transcripts to search or caption, if the bundled product does not already provide a transcript you can use.
What Sume does and does not do
Sume does not hold a live conversation or stream audio in either direction. Use it for the assets and records around an agent.
- Render fixed greetings, hold messages and legal notices once. See pre-rendered greetings.
- Transcribe recordings after the call at $0.01 a minute, in 10-minute jobs.
- Compare stack costs with the voice agent cost breakdown.
Sources
Related posts
More in Comparisons
- DeepL Voice keeps speakers' voices in 14 languages: Sume dubs
DeepL's September 15 release keeps each speaker's voice across 14 languages in live talk. A Sume dub picks a TTS voice id per line and does not clone.
- Describe a voice in words: Voice Design vs Sume Voices
ElevenLabs Voice Design takes 20 to 1000 characters. Sume Voices generate mode takes a 20 to 2000 character persona prompt and offers three variants.
- Does adding references help? A 3 versus 10 reference test plan on Sume
FLUX 3 Image takes ten references. Before you assume more is better, run 3 against 10 on your own products. A scoring sheet and script, no invented results.
- Does Sume have Veo 3.1? No. Which video models it lists instead
Sume's video catalog has no Veo model. See the models it does list, with duration and resolution ranges, and how Google's own docs now steer to Omni Flash.
Written by Sume