Aura-2 or Flux TTS: which Deepgram voice to pay for

Pay for Aura-2 at $0.030 per 1,000 characters when you are producing audio you will play back later, and for Flux TTS at $0.045 only when a live voice agent needs to know how much of a sentence the caller heard. Aura-1 at $0.015 is the budget tier. This choice comes from the price list and from trade press about Flux, so quality still needs a listening test.

The three tiers

Prices from Deepgram's pricing page; feature notes for Flux from audioXpress, which is trade press.

What Flux's extra 50% is for

A voice agent often gets interrupted. When the caller cuts in, the agent needs to know which part of its reply was spoken, so its memory of the conversation matches what the caller heard. According to audioXpress, Flux returns text_spoken and text_remaining for that case. The same article cites an 80 ms response latency claim and says the model launched English only; both are press-reported and neither was tested for this post.

Nothing in that feature set helps when no one interrupts. A narrated explainer, a product video or an ad read plays start to finish, so paying $0.015 more per thousand characters buys you nothing you can use.

Cost of one hour of narration

Cartesia's pricing page ties a minute of audio to 750 to 800 characters. Applying that ratio to Deepgram, which does not publish one, an hour of speech is about 45,000 to 48,000 characters. These are estimates; measure your own scripts.

A simple rule

Start every project on Aura-2 and move a line to Flux only when the product needs interruption tracking. Use Aura-1 for drafts and internal reviews where the voice is not the point.

For produced voiceover inside a video workflow, Sume offers Cartesia Sonic through its TTS Router, not Deepgram, with no streaming in the first version. Sume's Sonic-3.6 rate is $0.0475 per 1,000 characters before rounding up to the cent and the platform fee, which is above Flux's $0.045 on the listed numbers. The reason to choose it is the shared job flow, not the price.