ThunderPhone vs. Telnyx Voice AI Agents: the $0.05 rate, LLM tokens, telephony, and a 10,000-minute example
Telnyx and ThunderPhone bundle different parts of a voice agent. Telnyx combines orchestration, speech recognition, and voice, then bills language-model tokens and telephony separately. ThunderPhone includes the language model in its engine rate; carrier charges remain separate.
Decision summary: Telnyx suits teams that value its owned network and numbers, clean orchestration rate, and model selection. ThunderPhone suits teams seeking an all-in engine rate and documented regression tooling. Spark is ThunderPhone's lowest-priced engine. The pricing page describes Spark as "Somewhat slower latency, but acceptable for most use cases." ThunderPhone pricing. This is a cost comparison, not evidence of equal response speed or call quality.
Existing Telnyx customers can also keep their carrier and use ThunderPhone through the Telnyx integration. Choosing an agent platform does not require choosing a different number provider.
At-a-glance comparison
| Criterion | ThunderPhone | Telnyx Voice AI Agents | Qualification and source |
|---|---|---|---|
| Pricing model | Prepaid pay as you go. Spark 2¢/min, Bolt 5¢/min, Storm 9¢/min. | $0.05/min voice engine, plus LLM tokens and telephony. | Compare complete configurations. Telnyx pricing, accessed 2026-09-10. |
| What's included per minute | Speech recognition, language model, and voice. Premium paths and optional features cost extra. | Orchestration, STT, TTS, tools, and knowledge retrieval. LLM usage is separate. | The bundles have different boundaries. Telnyx pricing, accessed 2026-09-10. |
| Platform fees and concurrency | No subscription, seat, or platform fee; no published per-line concurrency fee. | Pay-as-you-go concurrency is 500 calls. No free-credit amount is stated in the reviewed pricing page; a card is required. | Other commitment terms are not verified here. Telnyx pricing, accessed 2026-09-10. |
| Languages | 47 languages, 123 voices; 15 included languages and 32 premium. Configured languages support mid-call switching. | A platform-wide language total is not published in the reviewed pricing material. | A larger-language-catalog claim is not supportable from these sources. |
| Telephony / BYO SIP | Telnyx numbers and other SIP trunks supported. No ThunderPhone import or monthly charge for imported numbers. | Own network and numbers; SIP inbound from $0.0032/min and outbound from $0.005/min. Number listed at $1/month. | Carrier charges remain payable with either agent. Telnyx SIP pricing and agent pricing, accessed 2026-09-10. |
| Compliance posture | GDPR and HIPAA compliant; no compliance add-on fee listed. SOC 2 is not claimed. | Compliance posture is not verified by the reviewed pricing evidence. | This is an evidence gap, not a claim that Telnyx lacks certifications. |
| Testing / monitoring tooling | Browser mic tests, simulations, regression gates, CI execution, and live-traffic A/B experiments. Simulations are billable real calls. | Comparable testing and monitoring details are not verified by the reviewed sources. | Assess the actual evaluation workflow before deciding. |
| Metering | Quarter-minute increments; 0.25-minute minimum. | General voice documentation specifies 60/60 billing. | Application to the agent-engine charge is unverified. Telnyx billing increments, accessed 2026-09-10. |
ThunderPhone figures and capabilities reflect its September verification. See ThunderPhone pricing, accessed 2026-09-10, for published engine and add-on terms.
How pricing actually compares
Use 2,000 calls × 5 connected minutes = 10,000 connected minutes/month, one primary included language, one phone number, and no more than 10 concurrent calls. Assume inbound US local traffic, with no transfers, hold, voicemail, recording, premium paths, or optional agent add-ons. This is illustrative planning arithmetic, not a quote.
What the Telnyx estimate means
Telnyx's voice-engine rate covers orchestration, STT, and TTS. Its separate Kimi LLM estimate is approximately $0.004/minute; its worked example uses a lower token-cost estimate. Neither is a fixed LLM tariff. Telnyx pricing, accessed 2026-09-10.
| Published input | Rate | Interpretation |
|---|---|---|
| Voice engine | $0.05/min | Orchestration, STT, and TTS. |
| Typical Kimi LLM estimate | Approximately $0.004/min | Telnyx's own estimate; actual usage bills per token. |
| LLM in Telnyx's worked example | Approximately $0.0032/min | Configuration-specific estimate used below. |
| Inbound SIP | From $0.0032/min | Starting carrier rate, subject to routing eligibility. |
| Outbound SIP | From $0.005/min | Different direction; excluded from this inbound example. |
| Telnyx's rounded worked total | Approximately $0.056/min | Includes engine, estimated LLM, and inbound telephony. |
Source for all inputs: Telnyx pricing, accessed 2026-09-10. Adding the displayed example components produces $0.0564/min, before number rental. That explains why exact component arithmetic differs from multiplying the rounded headline estimate.
Three configurations for the same workload
| Monthly component | Telnyx agent + Telnyx telephony | ThunderPhone Spark + Telnyx telephony | ThunderPhone Spark, telephony unpriced |
|---|---|---|---|
| Engine | 10,000 × $0.05 = $500 | 10,000 × $0.02 = $200 | 10,000 × $0.02 = $200 |
| Language model | 10,000 × estimated $0.0032 = $32 | Included in engine | Included in engine |
| Inbound carrier usage | 10,000 × starting $0.0032 = $32 | 10,000 × starting $0.0032 = $32 | Not priced |
| One carrier number | $1 | $1 | Not priced |
| Planning total | $565/month | $233/month | $200/month plus telephony and number rental |
Telnyx inputs: agent pricing and SIP pricing, accessed 2026-09-10. ThunderPhone engine input: pricing, accessed 2026-09-10. Totals are derived arithmetic, not published package prices.
The shared-carrier columns assume the starting inbound rate applies to each configured route, with no additional carrier services. Confirm that assumption before treating either as a budget. The final column deliberately leaves carrier usage and rental unpriced; its complete call cost is not computable without a telephony configuration.
Telnyx's rounded estimate would produce $560 in usage; the displayed component sum produces $564 before the number. The table uses the component sum consistently. It retains Telnyx's estimated token usage rather than assuming all models cost the same.
The existing Telnyx pricing explainer develops this same distinction. Its earlier credit and tier details are not reverified here; this comparison follows the September 10 verification.
Product differences that change the decision
Bundle clarity versus model selection
Telnyx's orchestration bundle is clear about what it includes. Selecting models independently can help teams tune behavior and costs, but introduces token-based forecasting.
| Kimi K2.6 token category | Published price per million tokens |
|---|---|
| Input | $0.665 |
| Cached input | $0.080 |
| Output | $4.000 |
Source: Telnyx inference pricing, accessed 2026-09-10. These prices do not establish a universal cost per conversation minute. Prompt length, cache usage, and generated output change the effective cost.
Anthropic and OpenAI pass-through token rates remain unverified. Telnyx labels those model options, but numerical rates were absent from the fetched agent and inference pages. Their configured costs are not computable from the reviewed public material.
ThunderPhone includes the language model in its engine rate and shows the configured all-in rate before launch. Here, all-in means the engine and selected ThunderPhone options; carrier billing remains separate. Long prompts and premium configurations can still increase the rate.
Keeping Telnyx numbers
Telnyx's network ownership and integrated numbers are practical reasons to choose its agent product. Teams already managing carrier operations there may prefer that consolidated setup.
Alternatively, connect existing numbers through ThunderPhone's Telnyx integration. ThunderPhone adds no import or monthly number charge, while Telnyx continues billing carrier usage and rental. Validate inbound routing, outbound requirements, tools, and transfer behavior before moving traffic.
Metering and additional charges
Telnyx documents 60/60 increments for voice calls. Whether that rule also applies to the voice-agent engine minute is unverified. ThunderPhone meters in quarter-minute increments with a quarter-minute minimum. The exact-length calls above avoid partial-minute rounding; shorter calls need a separate calculation. Telnyx billing increments, accessed 2026-09-10.
Telnyx separately lists Call Control and media-streaming charges, but their applicability to the agent route is unresolved in the reviewed sources. They are excluded from the table, following Telnyx's agent example. Confirm the chosen architecture against Call Control pricing, accessed 2026-09-10.
ThunderPhone's premium voice or language path adds up to 3¢/minute, non-stacking, or up to $300 for this workload. Hold time bills at 2¢/minute on every engine. Voicemail and phone-menu minutes use the engine rate; straight-to-voicemail calls are capped at one billable minute. See the ThunderPhone pricing guide before adding excluded call states.
Languages, compliance, and testing
ThunderPhone's language catalog and testing options are documented in its verified product information. Telnyx's corresponding language totals and evaluation features are not established by the reviewed pricing sources. Missing evidence should become a procurement question, not a feature-loss claim.
ThunderPhone is GDPR and HIPAA compliant; its published pricing lists no compliance add-on fee. SOC 2 is not claimed. Customers remain responsible for configuring their workflows, data handling, recordings, and notices for their own obligations.
Test the same caller scenarios before judging quality. ThunderPhone offers browser tests, AI-caller simulations, bot-to-bot and SIP-loopback calls, regression gates, CI execution, and live A/B experiments. Simulations are billable real calls. The pricing page describes Spark as "Somewhat slower latency, but acceptable for most use cases." ThunderPhone pricing. Test Spark against the intended conversation before choosing on price.
Who should choose which
Choose Telnyx when owning the carrier and agent relationship in one environment matters, or when explicit model selection supports your engineering workflow. Its clean orchestration rate makes the bundled layer easy to identify. Budget token usage and the configured carrier route separately.
Choose ThunderPhone when an LLM-inclusive engine rate and its documented regression workflow fit your priorities. Spark has lower modeled costs here. The pricing page describes Spark as "Somewhat slower latency, but acceptable for most use cases." ThunderPhone pricing. Keeping Telnyx telephony is a valid configuration, subject to confirming route charges.
For broader comparisons, read the cheapest AI voice-agent guide, voice AI pricing comparison, and voice AI price index. Build a deployment budget with AI phone-agent cost and the voice AI cost breakdown.
FAQ
Is Telnyx's $0.05 rate all-in?
It includes orchestration, STT, and TTS, but excludes LLM tokens and telephony. Telnyx's own worked example adds those components. Telnyx pricing, accessed 2026-09-10.
Which configuration costs less for this workload?
Spark with the modeled Telnyx route costs less in the table. That result assumes included-language settings, no optional add-ons, and carrier-rate eligibility. It does not establish equivalent response speed or quality.
Can I keep my Telnyx number with ThunderPhone?
Yes, ThunderPhone supports Telnyx numbers. Follow the Telnyx integration, retain carrier billing, and validate the intended call paths before migration.
Does Telnyx publish free voice-AI credits?
No free-credit amount is stated in the reviewed September pricing evidence. The agent page requires a card on file. Earlier offers are not verified here. Telnyx pricing, accessed 2026-09-10.
Is Telnyx's LLM estimate a fixed minute price?
No; actual LLM charges are token-based. The selected model, input, caching, and output affect cost. Frontier-model pass-through figures remain unverified. Telnyx agent pricing and inference pricing, accessed 2026-09-10.
Sources and freshness
- Telnyx Voice AI Agents pricing, accessed 2026-09-10.
- Telnyx inference pricing, accessed 2026-09-10.
- Telnyx Elastic SIP pricing, accessed 2026-09-10.
- Telnyx Call Control pricing, accessed 2026-09-10.
- Telnyx billing increments, accessed 2026-09-10.
- ThunderPhone pricing, accessed 2026-09-10.
This page reflects the September verification and is documentation-based: no account, invoice, or live call was tested. Unverified rates and feature details are identified above. Recheck route eligibility, token rates, and billing scope before purchase.