Deepgram Voice Agent API pricing (2026): rates, telephony, and the September changeover

Deepgram Voice Agent API charges for WebSocket connection time, with bundled speech and model options or bring-your-own components. Its Standard PAYG rate changes on a published schedule:

The API rate comes from Deepgram pricing; the floor adds Twilio inbound voice and Media Streams. Sources accessed 2026-09-10.

These are dated schedules, not an evergreen promotional price. Budget usage after September 12 using the later column. Telephony remains separate in either period.

How Deepgram's pricing model works

Standard and Advanced bundle orchestration, speech recognition, voice, and an included language model. Their model selections differ by tier. BYO LLM leaves your model bill outside Deepgram's rate.

Source: Deepgram pricing, accessed 2026-09-10. BYO TTS has no dated change listed; its displayed rate is repeated across periods.

Deepgram says Flux TTS in Voice Agent is free through September 12, 2026. That promotion does not make the entire agent free. Use the scheduled bundled rates after expiry; do not invent a separate Flux surcharge. Deepgram pricing, accessed 2026-09-10.

Growth advertises entry at $4K+/year and savings up to 20%. Those are plan terms, not an additional discount to subtract from the Growth rates above. Confirm commitment and credit terms before comparing annual spending with PAYG. Deepgram pricing, accessed 2026-09-10.

Advanced makes sense when your application needs its model selection. The tier label alone does not establish better results for your calls. Evaluate the relevant models against your workflow. Deepgram model documentation, accessed 2026-09-10.

What's included vs. billed separately

Sources: Deepgram pricing, telephony reference implementation, Twilio US pricing, and Twilio service pricing, accessed 2026-09-10.

Deepgram's inbound telephony documentation provides a Twilio reference implementation. This is a developer integration: the API handles the agent while Twilio supplies the phone connection. The calculation here uses Media Streams for that path. It does not add a separate ConversationRelay product charge.

Worked example: 10,000 minutes a month

Use 2,000 calls × 5 connected minutes = 10,000 connected minutes/month, one primary included language, one phone number, and no transfers, hold, or voicemail minutes. Assume no more than 10 concurrent calls; this is a workload assumption, not a verified Deepgram concurrency allowance.

This is illustrative planning arithmetic, not a quote. Assume WebSocket, carrier, and streaming durations each equal the connected workload. Credits, taxes, recording, optional services, and any rounding effects are excluded.

Arithmetic uses Deepgram pricing, Twilio US pricing, and Twilio service pricing, accessed 2026-09-10.

The per-minute floors are the API rate plus both Twilio usage rates. Number rental sits outside those floors. Each column assumes the entire workload receives that period's rate. For a month spanning the changeover, split connection minutes between the periods.

BYO LLM reduces the Deepgram line but leaves an unknown model bill:

These are derived from the same linked sources, accessed 2026-09-10. A complete BYO LLM total is not computable without model usage and applicable provider rates. A lower API line does not establish a lower complete bill.

Gotcha checklist

• Connection time is not talk time. Deepgram meters the open WebSocket. Silent periods can therefore count while it remains connected. Measure connection lifecycle alongside conversation duration.
• Telephony is outside the bundle. Include carrier minutes, streaming for this Twilio path, and number rental before calling a rate complete.
• The promotion expires. Preserve both dated columns in budgets. A trial before expiry does not establish the ongoing production rate.
• BYO is a separate bill. Model or voice provider costs can offset the lower Deepgram line.
• Rounding remains unresolved. Deepgram's fetched pricing does not state its increment. Twilio documents rounding by pricing model; the exact increment was not verified.
• Check capacity separately. The fetched Deepgram pricing page publishes no concurrency limit. Do not treat that absence as unlimited capacity.

Billing sources: Deepgram pricing and Twilio billing documentation, accessed 2026-09-10.

How ThunderPhone prices differ

ThunderPhone is prepaid pay as you go, without a subscription, seat, or platform fee. Each engine includes speech recognition, language model, and voice:

Source: ThunderPhone pricing, accessed 2026-09-10. Premium voice or language paths add up to 3¢/min, non-stacking. Metering uses quarter-minute increments with a 0.25-minute minimum.

Telephony is separate here too. Imported VoIP/SIP numbers carry no ThunderPhone import or monthly charge; your carrier bills usage. The builder shows the configured all-in rate before launch. Read ThunderPhone pricing explained for add-ons.

Deepgram's explicit BYO LLM and BYO TTS options suit teams choosing those components themselves. ThunderPhone suits buyers seeking bundled engines and phone-agent workflows. Engine-only figures should not be compared with Deepgram-plus-Twilio totals.

For broader comparisons, see the cheapest AI voice agent guide, voice AI pricing comparison, and voice AI price index. Use the cost breakdown and AI phone agent cost guide to build a complete budget.

FAQ

What does Deepgram Standard PAYG cost?

$0.056/min through September 12, 2026, then $0.075/min from September 13. These are API connection-minute rates, excluding telephony. Deepgram pricing, accessed 2026-09-10.

Is Deepgram Voice Agent free to try?

Deepgram advertises $200 in free credits, with no credit card required. Do not subtract that trial credit from recurring monthly projections or assume it pays Twilio charges. Deepgram telephony documentation, accessed 2026-09-10.

Does Deepgram charge only while someone speaks?

No; its published billing basis is WebSocket connection time. Speech duration alone can understate usage. The fetched pricing page does not specify rounding increments. Deepgram pricing, accessed 2026-09-10.

Does the Voice Agent API price include a phone number?

No; telephony is supplied separately. Deepgram documents a Twilio reference implementation, so budget the carrier and streaming path alongside the API. Deepgram telephony documentation, accessed 2026-09-10.

What does the Standard Twilio example cost monthly?

The computed floor is $690.15 before the changeover or $880.15 afterward for the stated workload. Both include number rental under the equal-duration assumptions above. Deepgram pricing, Twilio US pricing, and Media Streams pricing, accessed 2026-09-10.

Sources and freshness

• Deepgram pricing, accessed 2026-09-10.
• Deepgram Voice Agent model tiers, accessed 2026-09-10.
• Deepgram inbound telephony reference implementation, accessed 2026-09-10.
• Twilio US voice pricing, accessed 2026-09-10.
• Twilio voice service pricing, accessed 2026-09-10.
• Twilio usage and billing documentation, accessed 2026-09-10.
• ThunderPhone pricing, accessed 2026-09-10.

Every third-party figure on this page was checked against the linked first-party pages on September 10, 2026. It is documentation-based: no account, invoice, or live call was tested. Scheduled rates are as published on the verification date; later revisions require a fresh check. Deepgram rounding, concurrency, exact promo cutoff time zone, and complete BYO provider costs remain unverified or not computable from the fetched pages.