Deepgram Flux TTS Pricing: What It Costs and How to Budget It

Deepgram Flux TTS is free to use through September 12, 2026, as a launch promo. Because launch pricing changes, always confirm the current rate at deepgram.com/pricing before you build a budget around it.

Flux is Deepgram's conversation-native text-to-speech model, built for real-time voice agents. Text-to-speech is only one line item in a voice agent, so the price of the voice matters most when you see it next to the rest of the stack.

This guide explains the launch promo, how Deepgram prices text-to-speech today, and how per-character pricing turns into a real per-minute cost. Then it shows the full voice-agent cost picture and how to keep the bill low.

Is Deepgram Flux TTS Free Right Now?

Yes. Deepgram Flux TTS is free through September 12, 2026, as a launch promo. That gives teams a window to test it in a real voice agent before any usage charge starts.

Launch pricing is temporary by nature. A free tier during a launch tells you almost nothing about the long-term rate, so plan for a paid rate once the promo ends.

Confirm the live price on the vendor page before you commit. Deepgram lists current rates at deepgram.com/pricing.

How Does Deepgram Price Text-to-speech Today?

Deepgram prices text-to-speech by characters, not by the minute. Its general TTS model, Aura-2, is listed at $0.030 per 1,000 characters, which works out to roughly $0.02 to $0.03 per spoken minute.

Flux is a newer, separate model aimed at voice agents. Deepgram has not published a settled per-character Flux rate at the time of writing, so treat Aura-2's rate as a reference point, not a Flux quote.

Character-based pricing is common for text-to-speech. You pay for the text you send to be spoken, so shorter, tighter replies cost less than long, rambling ones.

How Do You Turn Per-character Pricing into a Per-minute Cost?

You estimate how many characters your agent speaks per minute, then multiply by the per-character rate. A normal speaking pace is about 150 words per minute, and an English word runs near 6 characters with spaces.

That gives you roughly 900 characters per spoken minute. Using Aura-2's published rate of $0.030 per 1,000 characters as the worked anchor, one spoken minute costs about $0.027.

Here is the method, step by step, so you can redo it with any rate you confirm on the vendor page:

This is Aura-2's rate used to show the method. Do not assume Flux charges the same. Once Deepgram publishes a Flux rate, drop it into the same formula.

The lesson: the words your agent speaks drive the bill. A wordy script costs more per call than a crisp one, even at the same rate.

• Step 1: estimate spoken words per minute (about 150 at a normal pace).
• Step 2: multiply by ~6 characters per word to get ~900 characters per minute.
• Step 3: divide by 1,000 and multiply by the per-1,000-character rate.
• Worked anchor (Aura-2 rate): 900 / 1,000 x $0.030 = about $0.027 per spoken minute.

What Does a Full Voice Agent Actually Cost?

Text-to-speech is one of three usage costs in a voice agent, so budget all three together. A chained voice agent runs speech-to-text, then a fast language model, then text-to-speech on every turn.

Speech-to-text listens to the caller. Deepgram's Nova streaming speech-to-text is listed near $0.0048 per minute. The language model in the middle decides what to say, and its cost depends on the model you pick.

Text-to-speech is the voice speaking back. Seen this way, the spoken voice is often a minority of the total per-minute cost once you add the caller's audio and the language model.

The table below shows the three layers so text-to-speech sits in context. Confirm every current rate on each vendor's own page.

Does Deploy-anywhere Change the Cost Math?

Yes. Deepgram Flux can run in your own cloud or on-premises, which trades per-call API fees for infrastructure you own and staff. That shift matters for teams with data-residency, security, or compliance rules.

Per-call API pricing is simple: you pay per character or per minute and Deepgram runs the servers. Self-hosting means you pay for GPUs, uptime, and the people who keep it running.

Self-hosting tends to pay off at high, steady volume, where fixed infrastructure beats a per-call rate. At low or spiky volume, the API is usually cheaper and far less work.

For exact self-host and enterprise terms, ask Deepgram directly and verify at deepgram.com/pricing.

How Does Flux Pricing Compare to ElevenLabs and Cartesia?

The pricing models differ more than the raw numbers. Deepgram charges per character with a deploy-anywhere option, ElevenLabs sells credit-based subscription tiers, and Cartesia prices its low-latency Sonic model on its own plans.

Comparing the shape of each model matters more than a single sticker price, because your usage pattern decides which shape is cheapest. The table below compares models, not invented numbers. Verify every rate on each vendor's page.

For a full breakdown of one rival's plans, see our guide to ElevenLabs pricing linked below.

How Do You Control Voice-agent Cost?

You control cost by trimming what the agent speaks and by measuring real call volume before you scale. Text-to-speech is billed on the words spoken, so tighter replies lower the bill directly.

Start with these levers, then re-run the per-minute math with rates you confirm on each vendor page:

When we evaluate a client's voice agent, the failure we see most is a voice that sounds warm on pickup and flat by the third turn, which is the exact problem conversation-native models target. Cost and quality both improve when the script is short and the model stays consistent.

• Shorten replies. Fewer characters spoken means a lower text-to-speech bill.
• Cache or template common answers so the agent does not regenerate them every call.
• Track average call length. Per-minute cost times minutes equals your real spend.
• Pilot on the free promo, then re-price against the confirmed post-promo rate.
• Pick a self-host or API path based on your actual volume, not a guess.