Voice agent, TTS API or rendered narration? Pick by output

Choose by the output you need: a live conversation calls for a voice agent such as Decagon Voice 3, a low-latency or high-fidelity speech API calls for MAI-Voice-2.1 or Gemini 3.8 TTS, and a finished video with narration, timings and captions calls for a job chain like Sume's. Four announcements in two weeks make this easy to confuse.

The four options in one table

Facts below come from each vendor's own page and Sume's rate card. Vendor prices are on their own pricing pages and are not repeated here.

Ask these three questions

Answer them in order.

• Is a person waiting for a reply right now? If yes, a voice agent.
• Do you only need speech from text, and will you place it yourself? Then a speech API.
• Do you need the speech joined to visuals, captions and a render? Then a job chain.

A chain on Sume

Sume's chain for a narrated clip is tts_create for the voice, timeline_audio to join parts, a caption job for burned text, and a timeline render. Each step is an async job with a job id, and jobs_wait returns the result. Captions reserve $0.20 for clips up to 60 seconds, audio detach is $0.01 per job and transcription is $0.01 per audio minute, per the rate card.

Mixing is normal

These are not substitutes. Check the shape of the output first, then compare prices on each vendor's pricing page.