Voice Support

Vital signs

Why it's on the table

On the table, Voice Support (Vs) is seat 37 of 58, in the Support & Success family. It is an experimental element — promising, volatile, and worth a contained experiment rather than a commitment. Budget curiosity, not dependence. It is optional: plenty of companies run without it — until a specific trigger (scale, regulation, cost, or customers) makes it essential for them. It sits in the mid price band — a real line item that should earn its keep visibly.

Voice Support: the top 5 — v2026.Q3

Edition v2026.Q3 · ranking, pricing and status verified 2026-08-06.

1. 1VapiVapi (YC W21)

   $0.05/min platform + models/voices at cost ($0 BYOK) · 10 concurrent lines incl., +$10/line/mo · HIPAA +$2k/mo, zero-retention +$1k/mo · enterprise volume-based

   Best for Developer teams who want full control of the voice stack — pick your own STT/LLM/TTS, test it, and ship a production support line as code.

   The developer default: 1B+ calls handled and enterprise revenue up 10x in the year to its $50M Series B (Peak XV, with M12 and Kleiner Perkins, May 12, 2026; $72M total). Amazon Ring routes 100% of inbound calls through it. The 2026 platform run — Monitoring (Apr), Model Intelligence with production-data model recommendations (Jul), agent test suites — is the deepest tooling in the category.

   Watch Orchestration-only means you own the debugging across four vendors' models. Add-ons bite: HIPAA is $2,000/mo before a single call. Ops teams without engineers ship faster on Retell or no-code tools. Series B valuation undisclosed.

   $50M Series B led by Peak XV, $72M total raised (May 12, 2026) [src] · 1B+ calls handled; enterprise revenue 10x YoY; Ring at 100% of inbound (May 2026) [src] · $0.05/min platform fee, models at cost with BYO keys (Aug 2026) [src]
   Site →Docs →

2. 2Retell AIRetell AI (YC W24)

   Pay-as-you-go $0.07–0.31/min all-in (infra $0.055 + TTS + LLM menu) · 20 concurrent incl. · $10 free credit · AI QA +$0.10/min · enterprise custom

   Best for Ops-led teams who want production support agents this week — dashboard-built, transparent per-minute menu pricing, monitoring and QA included.

   The capital-efficiency outlier: $60M ARR by April 2026 (650% YoY) on roughly $5M ever raised, running 40M+ real-time AI phone calls a month. Its 2026 releases — the Conductor agent runtime, MCP server, and the vCX-Hard benchmark (best model clears only ~88% of hard contact-center turns, Jun 2026) — show a team publishing honest evals of its own stack.

   Watch A ~$5M war chest against $11B-and-up rivals is a real asymmetry if the category turns capital-intensive. Component pricing stacks: knowledge base, PII redaction, and QA add-ons push the effective rate toward the top of the range. ARR and call-volume figures are Sacra-sourced, not company-published.

   $60M ARR, +650% YoY (Apr 2026); ~$45M end of 2025 [src] · 40M+ real-time AI phone calls per month; $4.6M seed (Aug 2024) is nearly all it has raised [src] · vCX-Hard benchmark: best of 27 models clears only ~88% of hardest contact-center turns (Jun 4, 2026) [src]
   Site →Docs →

3. 3ElevenLabs AgentsElevenLabs

   Free 15 min/mo · Starter $6/mo (75 min) · Creator $22 (275 min) · Pro $99 · Business $990 (12,375 min) · overage $0.08/min; LLM/telephony at cost

   Best for Teams for whom the voice itself is the product — best-in-class synthesis, 70+ languages, and the cheapest credible entry into phone support.

   The best voices in the business now bundled with an agent runtime: 4M+ agents deployed, phone/web/WhatsApp channels, and Eleven v3 Conversational (Feb 2026) tuned for turn-taking. Backed by an $11B company — $500M Series D, Feb 4, 2026 — so the model layer keeps compounding. At $0.08/min flat overage it undercuts most pure-plays.

   Watch The agents platform is younger than the pure-plays; contact-center depth (warm transfers, compliance workflows, QA tooling) trails Vapi and Retell. Company attention is split across music, dubbing, and TTS — support agents are one product among many. Case-study stats (83.4% resolution, 4.6/5 CSAT) are vendor-published single customers.

   4M+ agents deployed; 70+ languages, phone/web/WhatsApp/SMS (Aug 2026) [src] · $500M Series D at $11B valuation (Feb 4, 2026) [src] · $0.08/min overage on every tier; free tier 15 min/mo (Aug 2026) [src]
   Site →Docs →

4. 4Sierra (voice)Sierra (Bret Taylor)

   Outcome-based — pay per resolution, enterprise contracts only, no public rate card

   Best for Enterprises that want one agent across chat and phone, paid by resolved conversation rather than by the minute.

   Voice surpassed text as Sierra's primary channel by October 2025 — the strongest signal that phone is where enterprise agent spend is going. $200M ARR by May 2026 (from $26M end-2024), a $950M Series E at $15.8B (GV/Tiger, May 2026), and Rocket Mortgage, SiriusXM, Vanguard, and Gap on the line. Outcome pricing puts the failure-tolerance risk on the vendor.

   Watch No self-serve, no published pricing, enterprise sales cycle — startups get far more per dollar from Vapi or Retell. Outcome-based pricing only pencils at volume. Its chat side competes in element Sa; you're buying a suite, not a phone tool.

   $200M ARR (May 2026); $950M Series E at $15.8B valuation led by GV and Tiger Global [src] · Voice surpassed text as the primary interaction channel by Oct 2025 [src] · Outcome-based pricing: "only pay for the value Sierra delivers" (Aug 2026) [src]
   Site →Docs →

5. 5BlandBland AI

   Start $0.14/min (no card) · Build $0.12/min + $299/mo · Scale $0.11/min + $499/mo · enterprise custom, VPC/on-prem — flat all-in rate, no model pass-throughs

   Best for Regulated and high-volume teams who want one predictable per-minute price and a self-hosted, single-vendor stack.

   The vertically integrated bet: Bland runs its own STT/LLM/TTS (Bland Speech v3, Aug 2026) in its own infrastructure, so one flat rate covers everything — "no token charges, no surprise bills." $50M Series C (Jun 16, 2026; $100M+ total from Scale, Emergence, HubSpot, Dell) and 3.5M+ calls a week for Samsara, Kin Insurance, and CNO Financial.

   Watch Flat rate costs more at low volume than Vapi with BYO keys ($0.14 vs ~$0.06/min). Closed stack: you take Bland's models, good or bad, and can't swap in the frontier model of the month. Valuation and ARR undisclosed.

   $50M Series C, $100M+ total raised (Jun 16, 2026) [src] · 3.5M+ calls/week for Samsara, Kin Insurance, CNO Financial (Jun 2026) [src] · Flat $0.11–0.14/min bundling STT+LLM+TTS; VPC/on-prem at enterprise (Aug 2026) [src]
   Site →Docs →

Voice Support: the top 8 compared

Edition v2026.Q3 · ranking, pricing and status verified 2026-08-06.

How to choose your voice support

If engineers are building your phone line and you want to swap models as the frontier moves

Vapi — $0.05/min platform, bring your own keys, and the best test/monitoring tooling in the category.

If an ops-led team needs agents in production this week with a predictable all-in rate

Retell for the transparent per-minute menu; Bland if you'd rather one flat rate and a single vendor to blame.

If voice quality or multilingual coverage is the differentiator, or you're testing the channel on a budget

ElevenLabs Agents — 70+ languages, the best synthesis, and a $6/mo entry no pure-play matches.

If you're enterprise-scale, brand risk dominates, and you want the vendor to carry failure risk

Sierra's outcome-based pricing; in European or contact-center-first orgs, Parloa or PolyAI.

If you need latency and cost control at massive scale, or refuse platform lock-in

Build on LiveKit Agents or Pipecat — the open-source layer under half the platforms above — accepting you now own the whole pipeline.

Voice Support: the whole field

16 more tools tracked in this category, including 4 dead, renamed, or sunsetting — a reference that hides the graveyard isn't one. Verified 2026-08-06.

Voice Support: the category in numbers

Edition v2026.Q3 · ranking, pricing and status verified 2026-08-06.

• AI voice agents market ~$3.5B (2026) → $35.2B by 2033, 39.0% CAGR [src]
• Voice overtook text as Sierra's primary channel by Oct 2025; Sierra hit $200M ARR and a $15.8B valuation ($950M Series E) by May 2026 — enterprise agent spend is moving to the phone [src]
• 2026 funding wave across the whole stack: ElevenLabs $500M at $11B (Feb 4), Parloa $350M at $3B (Jan 15), Vapi $50M (May 12), Bland $50M (Jun 16), PolyAI $86M (Dec 15, 2025) [src]
• Retell reached $60M ARR (Apr 2026, +650% YoY) on ~$5M raised — voice-agent platforms can scale revenue without scaling capital [src]
• Consolidation began below the platforms: NICE acquired Cognigy (~$955M, Jun 2025); Salesforce absorbed Tenyx (Sep 2024); model layer commoditizing — gpt-realtime GA Aug 28, 2025 at 20% below preview pricing [src]
• Quality ceiling is real: Retell's vCX-Hard benchmark shows the best of 27 frontier models clears only ~88% of the hardest real contact-center turns (Jun 2026) [src]

Voice Support: method & sources

Scope: TTS/voice generation is element Vo — ElevenLabs appears in both tables and is judged here only on its Agents runtime. Chat-first support agents (Fin, Decagon — both added voice in 2025–26) stay in Sa; Sierra is ranked here because voice is now its primary channel by its own data, but it straddles both elements. Outbound sales dialers (Thoughtly, Air AI's ghost) belong closer to the outbound/sales elements. Ranking criteria: production call volume, pricing transparency, developer/ops fit for startups, and enterprise ceiling — in that order. Conflicts and weak spots: Retell's ARR and 40M calls/mo are Sacra-only (company doesn't publish); Cognigy's ~$955M price is widely reported but the primary NICE release was unreachable; Vapi and Bland valuations are undisclosed; PolyAI's 391% ROI is a vendor-commissioned Forrester study; Parloa's own press page is stale (lists only the Series B) — Series C/D figures verified via EU-Startups; Tenyx acquisition details are from general reporting, not re-verified primary sources. Air AI caveat: the air.ai domain was taken over by Govini's defense rebrand in 2026 — do not confuse the two. Ranking criteria: verified commercial traction, independent satisfaction surveys, agent benchmarks, and founder-fit (price floor, lock-in, surfaces). Editorial, never paid — the charter. Machine-readable twin: vs.json.

Our take

Real and improving, but tolerance for failure is lower on the phone. Deploy behind chat, not before it.

Combines with