Deepgram Flux TTS: $0.045 per 1K characters after the free period
Deepgram's Flux TTS is listed at $0.045 per 1,000 characters, 1.5 times Aura-2. Trade press says the free period ended Sept 12 and launch was English only.

Deepgram lists Flux TTS at $0.045 per 1,000 characters, which is $45 per million. Trade press reports that Flux TTS launched on 2026-08-13 with a free tier through 2026-09-12 and standard pricing from 2026-09-13, so the free period is over. The pricing page corroborates the price; the launch dates and the English-only launch come from audioXpress and are press-reported.
The listed price
Deepgram's pricing page lists three text-to-speech models on 2026-10-05.
| Model | Per 1,000 characters | Per 1M characters | Versus Aura-2 |
|---|---|---|---|
| Aura-1 | $0.015 | $15 | 0.5x |
| Aura-2 | $0.030 | $30 | 1.0x |
| Flux TTS | $0.045 | $45 | 1.5x |
What press coverage adds
From audioXpress, trade press. Treat these as reported, not as Deepgram's own statements.
- Flux TTS launched 2026-08-13 as a conversation-aware model for voice agents.
- An 80 ms response latency was claimed. It is a vendor claim relayed by the press and was not measured for this post.
- On an interruption it returns
text_spokenandtext_remaining, so an agent knows how much of its sentence the caller heard. - English only at launch.
- Free through 2026-09-12; standard price from 2026-09-13.
What it costs for real scripts
Arithmetic on the listed rates. Cartesia's page says a minute of audio takes 750 to 800 characters, which this post uses only to translate characters into minutes; Deepgram does not state a ratio, so the minute column is an estimate.
| Script | Flux TTS | Aura-2 | Approx. audio at 750-800 characters a minute |
|---|---|---|---|
| 1,000 characters | $0.045 | $0.030 | about 1.25 to 1.33 minutes |
| 10,000 characters | $0.45 | $0.30 | about 12.5 to 13.3 minutes |
| 1,000,000 characters | $45 | $30 | about 21 to 22 hours |
When Flux is worth 1.5x
The extra $0.015 per thousand characters buys the interruption bookkeeping and the low latency claim, which matter for live voice agents that must stop mid-sentence and resume correctly. For produced audio, such as an ad read or a narrated video, those features do nothing, and Aura-2 or another engine is the better value. Many teams will pay for Flux only on the live path.
Sume does not ship Deepgram. Its TTS Router offers Cartesia Sonic, billed at $0.0475 per 1,000 characters (Cartesia's $38 per million list times 1.25) before rounding up to the cent and the 5.5% platform fee, and it has no streaming in its first version, so it suits produced audio rather than live agents.
Sources
Related posts
More in Pricing
- Do Sume wallet credits expire? 365 days per top-up lot
Each Sume top-up becomes its own credit lot with a 365-day expiry. How to size a top-up so nothing lapses, and how that differs from monthly credits.
- Does a Sume 402 insufficient_credits cost you? No, nothing ran
A 402 on submit means Sume could not reserve the estimate. No provider work started and nothing is captured. Three ways to shrink the request and retry.
- 160 earnings briefs a year at 3,200 characters: TTS cost, MAI vs Sume
Forty companies, four quarters, 3,200 characters each is 512,000 characters: $7.68 on MAI Flash, $11.26 on MAI-Voice-2.1 and $24.32 on Sume TTS.
- Eight low-quality edits cost less than one high GPT Image 2.5 edit
On Sume, a GPT Image 2.5 edit with one reference costs $0.0094 at low and $0.0835 at high, so eight low edits cost $0.075, less than one high edit.
Written by Sume