A8gent
A8gent
·
Pricing

Twilio Voice Agent Pricing: Eight Meters, One Min

ConversationRelay is $0.07 a minute and that is the smallest line on the bill. The full Twilio conversational AI stack, meter by meter, with the LLM leg that is not in Twilio's table.

Twilio Voice Agent Pricing: Eight Meters, One Min visual
AI Agent Starter Kit cover

Free download

AI Agent Starter Kit

Workflow picker, ROI worksheet, 40-prompt pack, and the first-agent rollout playbook. The 1-hour version of the entire A8gent system.

Workflow picker
ROI worksheet
Prompt checklist
Team rollout plan
Open it now

No spam. Worksheet + prompt checklist + rollout plan delivered instantly.

Contents
  1. Fast answer
  2. Quick facts
  3. What this covers
  4. Why it matters
  5. Where to go next
  6. Outcomes
  7. Step by step
  8. Every meter that fires on one minute of conversation
  9. Conversation Memory is three meters wearing one name
  10. The largest line is the one Twilio does not bill you for
  11. At a glance
  12. Mistakes to avoid
  13. FAQ
  14. Related guides
TL;DRAnswer-first

Fast answer

Twilio publishes its conversational AI rates, which makes this the opposite problem to the enterprise vendors: the numbers are all there and nobody adds them up. ConversationRelay is $0.07 per minute, and Twilio states voice costs are calculated separately. Conversation Orchestrator is $0.0002 per 1,000 characters ingested. Conversation Intelligence is $0.005 per 1,000 characters for Twilio-authored language operators, or $0.002 per 1,000 input characters plus $0.018 per 1,000 output characters for custom operators. Conversation Memory carries three meters at once: $0.005 per daily utilized profile, $0.0028 per 1,000 characters of memory generation, and $0.007 per recall, with the first 100,000 profile operations free each month and $0.08 per 1,000 after that. Enterprise Knowledge adds $0.018 per GB-hour of storage plus $0.005 per retrieval. Every one of those can be live on the same minute, and your own LLM tokens are not on Twilio's table at all. Rates read 1 September 2026.

Facts last verified 2026-09

Quick facts

ConversationRelay
$0.07/min; voice costs calculated separately (checked 2026-09-01)
Conversation Orchestrator
$0.0002 per 1K characters ingested
Conversation Intelligence, Twilio operators
$0.005 per 1K characters, input and output combined
Conversation Intelligence, custom operators
$0.002 per 1K input characters, $0.018 per 1K output characters
Conversation Memory, profiles
$0.005 per daily utilized profile
Conversation Memory, generation
$0.0028 per 1K characters, input and output
Conversation Memory, recall
$0.007 per recall
Profile operations
First 100,000 free monthly, then $0.08 per 1,000
Enterprise Knowledge
$0.018 per GB-hour storage, $0.005 per retrieval
Language model tokens
Not billed by Twilio; your model provider charges separately

On this page

What this page covers

A comparison visitor should understand the tradeoff, the best-fit scenario, and the next diagnostic tool to confirm the choice.

  1. 01What Twilio publishes
  2. 02The meters that stack on one minute
  3. 03Conversation Memory has three meters
  4. 04The LLM leg is yours
  5. 05A worked per-minute stack
  6. 06When a bundled platform wins

Why does this matter now?

The $0.07 ConversationRelay minute is quoted constantly and it is one line of a bill with up to eight. A voice agent that remembers callers, is orchestrated across channels, is analysed for intent and grounded in a knowledge base touches five Twilio products and one model provider that is not Twilio. None of that is hidden, but it is spread across a pricing page in units that do not compose: minutes, characters, profiles, recalls, GB-hours and retrievals. Adding them up in your own units before you build is the difference between a composed stack that is cheaper than a bundled platform and one that quietly is not.

Internal path

Where to go next from this page

These links are part of the A8gent learning and conversion path. Use them to move from concept, to diagnosis, to workflow build, to course.

Start with readiness

What you should be able to do after this

  • Add up every meter a voice minute touches
  • See which costs Twilio does not bill and you still pay
  • Understand the three separate Conversation Memory meters
  • Compare a composed stack against a bundled platform

How do you do it, step by step?

1. Start with the relay minute and admit it is not the bill

ConversationRelay is $0.07 per minute for the real-time bridge between the call and your agent. Twilio states voice costs are calculated separately, so the telephony leg, the numbers and the carrier minutes are a different line item on the same invoice. Treat $0.07 as the floor and never as the estimate.

2. Convert your conversation into characters, twice

Orchestrator and Intelligence both meter in 1,000-character blocks and they are not the same rate. Orchestrator ingests at $0.0002 per 1,000 characters. Intelligence with Twilio-authored operators is $0.005 per 1,000 characters across input and output combined, while custom operators split into $0.002 per 1,000 input and $0.018 per 1,000 output. Custom output at $0.018 is over three and a half times Twilio's blended combined rate, which is the steepest step on the page.

3. Count all three Conversation Memory meters

Memory is not one price. It is $0.005 per daily utilized profile, $0.0028 per 1,000 characters of memory generation counting input and output, and $0.007 per recall. Profile operations are free for the first 100,000 a month and $0.08 per 1,000 thereafter. A design that recalls memory several times per conversation multiplies the recall meter by turn count, not by call count.

4. Add Enterprise Knowledge if you are grounding answers

Retrieval-grounded answers add $0.018 per GB-hour of storage plus $0.005 per retrieval. Storage is billed by time rather than by use, so a large indexed corpus costs money on a day with no calls, while retrieval scales with how often your agent looks something up, which is usually more than once per conversation.

5. Add your own model bill, which Twilio does not show

ConversationRelay bridges to your language model; it does not include it. The tokens your agent consumes are billed by your model provider at their rates and appear nowhere in Twilio's pricing table. On a chatty multi-turn call this is frequently the largest single component, and leaving it out is the most common reason a composed stack looks cheaper than a bundled one on paper.

Every meter that fires on one minute of conversation

Twilio's conversational AI pricing page is honest and complete, and it is still misread constantly, because the headline number is the smallest one on it. ConversationRelay is $0.07 per minute, and Twilio states directly on the same page that voice costs are calculated separately. That is the bridge between the phone call and your agent. It is not the phone call, it is not the transcription, it is not the reasoning, and it is not the memory. It is one component of a stack that most production voice agents assemble from five Twilio products plus a model provider Twilio does not bill for.

Conversation Orchestrator is $0.0002 per 1,000 characters ingested from conversations. That is a small number and it stays small, which makes it the least interesting meter on the page and the easiest to leave out of a model without consequence. Conversation Intelligence is where the character-based meters start to matter: $0.005 per 1,000 characters for Twilio-authored language operators, counting input and output combined, or a split rate for custom operators of $0.002 per 1,000 input characters and $0.018 per 1,000 output characters.

That custom-operator output rate deserves attention because it is the steepest step anywhere on the page. At $0.018 per 1,000 output characters, custom operators cost more than three and a half times the blended Twilio-operator rate of $0.005, and the input side at $0.002 is cheaper, so the switch to custom operators shifts cost onto whichever side of the exchange produces the most text. Teams reach for custom operators because they want a bespoke extraction or classification, which is a perfectly good reason, but the pricing consequence is not visible unless you notice that the meter changed shape from combined to split when you switched.

Enterprise Knowledge is the last of the straightforward meters and it has an unusual property: half of it is billed by time rather than by use. Storage is $0.018 per GB-hour and retrieval is $0.005 each. Storage accrues on a quiet Sunday exactly as it does during a peak hour, so a large indexed corpus has a floor cost independent of call volume. Retrieval scales with how often your agent looks something up, which in a grounded agent is usually more than once per conversation and sometimes once per turn.

Conversation Memory is three meters wearing one name

If there is a single line item on this page that breaks forecasts, it is memory, because people budget it as one price and it is billed as three. The published rates are $0.005 per daily utilized profile, $0.0028 per 1,000 characters of memory generation counting input and output together, and $0.007 per recall. There is also a fourth number attached: profile operations are free for the first 100,000 each month and then $0.08 per 1,000 operations.

Each of those three meters scales on a different axis, which is why they cannot be collapsed into one estimate. The profile charge scales with distinct people who interacted on a given day, so it tracks your customer base rather than your call volume. Memory generation scales with how much text gets written into memory, which tracks conversation length. Recall scales with how many times the agent reaches back for context, which tracks conversational design rather than either of the other two. A verbose agent that recalls on every turn hits all three axes at once.

The recall meter is the one that surprises people, and the surprise is structural. At $0.007 a recall the unit looks trivially cheap next to a $0.07 minute, and the instinct is to ignore it. But recall is a per-turn behaviour in most agent designs rather than a per-call one, so a ten-turn conversation that recalls context each turn is ten recalls, which is $0.07, which is exactly the relay minute you were treating as the headline. One design decision has doubled a line you thought you understood.

The practical response is not to avoid memory, which is often the feature that makes a voice agent worth deploying. It is to instrument the three meters separately from day one, because you cannot optimise an aggregate. Know your daily utilized profile count, your characters generated per conversation, and your recalls per conversation as three distinct numbers. Once they are separate, each has an obvious lever: profile hygiene, output constraints, and recall-on-demand rather than recall-every-turn.

The largest line is the one Twilio does not bill you for

ConversationRelay connects the call to a language model you supply and operate. Your model provider bills you for those tokens at its own rates, and none of that appears on Twilio's conversational AI pricing table. This is not concealment; it is the architectural point of the product, which exists precisely so you can bring your own model. But it means the total cost of a Twilio voice agent cannot be read off Twilio's pricing page at all, and on a chatty multi-turn call the model leg is frequently the largest single component of the total.

Add the telephony leg on the same principle. Twilio states that voice minutes, SMS messages, WhatsApp conversations and other channel costs are billed separately, and that voice conversations must be transcribed before Conversation Orchestrator can capture them, with transcription available through the Twilio Voice API. So a single minute of a grounded, remembering, orchestrated voice agent can touch the relay rate, the carrier rate, transcription, orchestrator ingestion, intelligence characters, three memory meters, knowledge storage and retrieval, and your own token bill. That is the honest picture, and it is why the $0.07 figure quoted in isolation is close to meaningless.

None of this makes Twilio expensive. Composed stacks exist because they let you swap the model when a cheaper one will do, tune transcription for your accents, and keep the telephony relationship you already have. Those are real advantages that a bundled per-minute platform cannot offer at any price, and for a team with engineers they are frequently worth more than the arithmetic. The mistake is not choosing Twilio; it is choosing it on a per-minute comparison against platforms whose per-minute rate includes things Twilio's does not.

So run the comparison in matched units. Bland, Retell, ElevenLabs Agents and LiveKit Agents all publish per-minute or per-plan rates with different bundling boundaries, and each of our pricing pages for those platforms states what sits outside the bundle. Build your Twilio stack total, model tokens and telephony included, then put it next to a bundled rate with the same components counted. Whichever way that comes out, it will be a comparison of two real numbers instead of a comparison of a total against a component.

At a glance

Twilio conversational AI published rates, read from twilio.com on 1 September 2026

ProductMeterPublished rate
ConversationRelayPer minute of relayed conversation$0.07/min, voice billed separately
Conversation OrchestratorCharacters ingested from conversations$0.0002 per 1K characters
Conversation IntelligenceTwilio-authored language operators$0.005 per 1K characters, input plus output
Conversation IntelligenceCustom language operators, input$0.002 per 1K characters
Conversation IntelligenceCustom language operators, output$0.018 per 1K characters
Conversation MemoryDaily utilized profile$0.005 per profile per day
Conversation MemoryProfile operationsFirst 100K free monthly, then $0.08 per 1K
Conversation MemoryMemory generation$0.0028 per 1K characters, input plus output
Conversation MemoryRecall$0.007 per recall
Enterprise KnowledgeStorage$0.018 per GB-hour
Enterprise KnowledgeRetrieval$0.005 per retrieval
Voice, SMS, WhatsApp channelsCarrier and channel usageBilled separately, not on this page
Voice transcriptionRequired before Orchestrator captureBilled through the Twilio Voice API
Language model tokensYour own model providerNot billed by Twilio

What mistakes should you avoid?

  • Quoting $0.07 a minute as the cost of a Twilio voice agent
  • Forgetting the voice and telephony leg, which Twilio bills separately
  • Treating Conversation Memory as one meter when it is three
  • Using the Twilio-operator Intelligence rate when you are running custom operators, whose output rate is $0.018 per 1K characters
  • Counting recalls per call when a multi-turn agent recalls per turn
  • Leaving your own LLM token spend out of the comparison entirely

FAQ

How much does Twilio ConversationRelay cost?
$0.07 per minute as published on 1 September 2026, with voice costs calculated separately. That covers the real-time relay between the call and your agent only. It does not include telephony, transcription, orchestration, intelligence, memory, knowledge retrieval or the language model itself.
What are all the Twilio conversational AI meters?
ConversationRelay at $0.07 per minute, Conversation Orchestrator at $0.0002 per 1,000 characters, Conversation Intelligence at $0.005 per 1,000 characters for Twilio operators or $0.002 input and $0.018 output per 1,000 characters for custom, Conversation Memory at $0.005 per daily utilized profile plus $0.0028 per 1,000 characters generated plus $0.007 per recall, and Enterprise Knowledge at $0.018 per GB-hour plus $0.005 per retrieval.
Does Twilio include the LLM in its voice agent pricing?
No. ConversationRelay connects the call to a model you supply and operate, and your model provider bills you for tokens at its own rates. Those charges do not appear anywhere on Twilio's conversational AI pricing page, so any comparison against a bundled voice platform has to add them back in.
Are Twilio voice minutes included in ConversationRelay?
No. Twilio states that voice costs are calculated separately, and that voice conversations must be transcribed before Conversation Orchestrator can capture them, with transcription billed through the Twilio Voice API. Budget the carrier leg and transcription as their own line items.
Is Twilio cheaper than a bundled voice agent platform?
Only if you add up every meter first. A bundled platform quotes one per-minute rate that includes the model, transcription and voices. A Twilio stack quotes $0.07 plus telephony plus transcription plus your token bill plus whichever of Orchestrator, Intelligence, Memory and Knowledge you enabled. The composed stack wins on control and on tuning headroom, not automatically on price.

Sources & further reading

Done for you

Want this built for your business?

Tell us the workflow. We scope, build, and ship the agent with guardrails and the numbers to prove it worked. The scoping call is free.

Free scoping call. You own the code.

Request a scoping call

Was this page helpful?