Vapi vs Retell vs Bland: Real Voice-Agent Cost per Connected Minute (2026)
Short answer: The advertised rate on each platform's homepage measures three different things, not the same thing. Vapi's $0.05/min is an orchestration fee only, with speech-to-text, the LLM, text-to-speech, and telephony billed separately at cost. Retell's own pricing page quotes a configuration-dependent range of $0.07–$0.31/min for its Pay-as-you-go tier — not a single bundled rate — where $0.07 is the floor (its $0.055/min voice-infrastructure charge plus a $0.015/min standard TTS voice, before telephony or a meaningful LLM cost is added) and $0.31 is the ceiling once you add a premium LLM and premium TTS voice. Bland is the only one of the three whose current public self-serve pricing bundles all four components (LLM, STT, TTS, telephony) into one flat, plan-based rate — and its current live pricing page shows only two self-serve tiers, Start and Build, plus custom Enterprise; the "Scale" tier ($499/month, $0.11/min) that some 2026 write-ups still reference is no longer shown on Bland's own pricing page as of September 26, 2026. Normalized to a real assembled stack, a lean configuration on Retell is the cheapest option at every volume we modeled; Bland's flat rate becomes more competitive as the stack gets more expensive (premium LLM, premium voice), and at very high volume a real buyer would likely be quoted Bland's custom Enterprise rate rather than paying Build's public per-minute price indefinitely.
Pricing at a glance
| Vapi | Retell | Bland | |
|---|---|---|---|
| Billing model | Orchestration fee + pass-through | Componentized, configuration-dependent | Fully bundled, plan-based (self-serve tiers: Start, Build; Enterprise custom) |
| Platform/voice-layer rate | $0.05/min (orchestration only) | $0.055/min voice infrastructure + $0.015/min (standard TTS) or $0.040/min (ElevenLabs TTS) | Start: $0.14/min talk time (free plan) · Build: $0.12/min ($299/month platform fee) · Enterprise: custom |
| Official all-in range | N/A (fully modular) | $0.07–$0.31/min, per Retell's own pricing page, depending on LLM and TTS choice | Start/Build rates above are already all-in (LLM, STT, TTS included, no separate charges) |
| STT | Billed separately at cost (e.g., Deepgram Nova ≈ $0.004/min), or $0 with your own API key | Included in the voice-infrastructure rate | Included |
| LLM | Billed separately at cost; ranges from ~$0.003/min (lightweight models) to $0.16–$0.32/min (top-tier models, standard vs. "fast" tier) | Same wide range as Vapi, billed per-minute by model choice (e.g., GPT-5 nano ≈ $0.003/min up to GPT-5.5 at $0.16–$0.32/min) | Included in the flat rate (model choice more limited) |
| TTS | Billed separately at cost; standard voices ~$0.01–0.03/min, premium cloned voices meaningfully more | $0.015/min (Retell Platform, Minimax, Fish, Cartesia, OpenAI voices) or $0.040/min (ElevenLabs voices) | Included |
| Telephony | Pass-through (e.g., Twilio ~$0.014/min), or $0 bring-your-own | $0.015/min via Retell's own telephony (US), or $0 bring-your-own | Included; transfer time to a human billed separately ($0.04–$0.05/min depending on plan) |
| Concurrency | 4 concurrent calls included on the self-serve/usage-only tier; $10/month per additional line | 20 concurrent calls included, more available on demand (confirmed current); additional concurrency $8/month per line beyond the first 20 | Start: 10 concurrent calls · Build: 50 concurrent calls · Enterprise: sized to volume |
| Compliance | HIPAA add-on ~$2,000/month; Zero Data Retention ~$1,000/month | Not separately broken out in public pricing | Not separately broken out in public self-serve pricing |
Sources: Vapi's own pricing page (vapi.ai/pricing) and billing documentation (docs.vapi.ai/billing/pricing-and-success-packages, docs.vapi.ai/billing/purchase-call-concurrency); Retell's own pricing page (retellai.com/pricing) with its live component-pricing calculator; Bland's own pricing page (bland.ai/pricing), directly fetched on this research date.
Why the headline rate is never the real rate
Vapi sells an orchestration layer, not a finished voice agent. The $0.05/min buys real-time pipeline management, barge-in handling, call routing, transcripts, and observability — nothing else. Every other component is charged "at cost," meaning Vapi adds no markup, but you're still paying it, either through Vapi's billing or directly to the provider if you bring your own API keys. Vapi's default self-serve concurrency is lower than commonly assumed: its own billing documentation lists 4 concurrent call lines on the base usage-only tier, not 10 — 10 concurrent lines is actually the limit on Vapi's paid "Core" success package, a separate add-on tier with its own pricing, not the self-serve default.
Retell does not sell a single bundled rate. Its own pricing page presents a live calculator and a stated range of $0.07–$0.31/min because the voice-infrastructure charge ($0.055/min) and TTS charge ($0.015–$0.040/min depending on voice) are fixed, but the LLM charge varies enormously by model — from about $0.003/min for a lightweight model up to $0.16–$0.32/min for a top-tier model at standard vs. "fast" processing tiers. Treating "$0.07/min" as Retell's rate, full stop, understates the real range its own pricing page publishes.
Bland is the outlier: its current public self-serve rate is genuinely all-inclusive for the core call, with no separate STT/LLM/TTS bills. The tradeoffs are less model choice and less pipeline customization than the other two, and — as of September 26, 2026 — only two self-serve tiers rather than three.
Hidden costs to model before you buy
- Vapi's real default concurrency is 4, not 10. A team that assumes 10 free concurrent lines (a figure still widely repeated in older write-ups) will hit Vapi's concurrency ceiling, and the resulting queuing or dropped calls, sooner than expected. Getting to 10 concurrent lines either means paying $10/month for each of 6 additional lines on the base tier, or upgrading to Vapi's "Core" success package, which bundles 10 concurrent lines with a different support tier at its own separate price.
- HIPAA compliance is not included on Vapi's base rate and adds roughly $2,000/month regardless of call volume — a fixed cost that only makes sense to absorb once volume is high enough to amortize it.
- Retell's $0.07 floor assumes a near-zero-cost LLM and no telephony charge. The moment you add Retell's own telephony ($0.015/min) and any real LLM (even a cheap one, ~$0.003/min), the realistic floor moves to roughly $0.088/min — still well under the $0.07–$0.31 range's midpoint, but meaningfully above the bare $0.07 figure often quoted in isolation.
- Retell's concurrency overage is priced per line, not as a step function — $8/month for each concurrent call line beyond the first 20, which matters for a team running frequent simultaneous bursts rather than steady, spread-out call volume.
- Bland's transfer time is billed separately from connected-call time, at $0.04/min (Build) to $0.05/min (Start) when using Bland-provided numbers (free with your own Twilio/BYOT).
- Bland no longer publishes a mid-tier self-serve plan between Build and Enterprise. A team outgrowing Build's 50-concurrent-call, $0.12/min terms has no public next step to check pricing against — the next tier is a custom Enterprise quote, which means the previously-published "Scale" numbers ($499/month, $0.11/min) should not be assumed to still apply.
Worked scenarios
We model two stack tiers on Vapi and Retell — a lean stack (small/open LLM, standard TTS voice, standard telephony) and a premium stack (a capable-to-top-tier LLM, premium TTS voice) — against Bland's flat, plan-based rate, using only the tiers currently shown on Bland's own pricing page (Start and Build; Enterprise is custom and not quantifiable here).
Assumptions: Vapi lean pass-through ≈ $0.063/min (STT ~$0.004 + LLM ~$0.015 + TTS ~$0.03 + telephony ~$0.014); Vapi premium pass-through ≈ $0.148/min (STT ~$0.004 + LLM ~$0.05 + TTS ~$0.08 + telephony ~$0.014) — these are illustrative, provider-choice-dependent figures, not official Vapi rates for a specific bundle. Retell lean uses its own published component rates directly: $0.055 (voice infra) + $0.015 (standard TTS) + $0.003 (lightweight LLM) + $0.015 (Retell telephony) = $0.088/min. Retell premium uses the official ceiling of its own published range, $0.31/min, rather than an assumed blend.
1,000 connected minutes/month
| Platform | Config | Cost | Per-minute |
|---|---|---|---|
| Vapi | Lean | $50 platform + $63 pass-through = $113 | $0.113 |
| Vapi | Premium | $50 platform + $148 pass-through = $198 | $0.198 |
| Retell | Lean | 1,000 × $0.088 = $88 | $0.088 |
| Retell | Premium | 1,000 × $0.31 (Retell's own published ceiling) = $310 | $0.31 |
| Bland | Start (no monthly fee) | 1,000 × $0.14 = $140 | $0.14 |
At this volume, Retell's lean stack is cheapest ($88), followed by Vapi's lean stack ($113) and Bland Start ($140). Bland Build's $299 monthly fee doesn't pay for itself yet at this volume.
10,000 connected minutes/month
| Platform | Config | Cost | Per-minute |
|---|---|---|---|
| Vapi | Lean | $500 + $630 = $1,130 | $0.113 |
| Vapi | Premium | $500 + $1,480 = $1,980 | $0.198 |
| Retell | Lean | $880 | $0.088 |
| Retell | Premium | $3,100 | $0.31 |
| Bland | Start | $1,400 | $0.14 |
| Bland | Build ($299/mo) | $299 + $1,200 = $1,499 | $0.15 |
Retell's lean stack remains cheapest ($880); Bland's Start plan is still more economical than Build at this volume — the fixed fee hasn't paid for itself yet.
50,000 connected minutes/month
| Platform | Config | Cost | Per-minute |
|---|---|---|---|
| Vapi | Lean | $2,500 + $3,150 = $5,650 | $0.113 |
| Vapi | Premium | $2,500 + $7,400 = $9,900 | $0.198 |
| Retell | Lean | $4,400 | $0.088 |
| Retell | Premium | $15,500 | $0.31 |
| Bland | Build ($299/mo) | $299 + $6,000 = $6,299 | $0.126 |
Retell's lean stack is still the cheapest fully-quantifiable option ($4,400). Bland's Build plan ($6,299) is now the only confirmed self-serve Bland number at this volume — its public pricing page offers no higher self-serve tier, so a real buyer at 50,000 minutes/month should expect Bland to quote a custom Enterprise rate rather than staying on Build. Bland does not publish an Enterprise rate; older third-party figures referencing a sub-$0.11/min enterprise rate are not verifiable against Bland's current published pricing and should be treated as unconfirmed.
Break-even: when does a platform fee pay for itself?
- Bland Build vs. Start: the $299/month fee saves $0.02/min ($0.14 → $0.12); it pays for itself above roughly 15,000 connected minutes/month ($299 ÷ $0.02). There is currently no further published self-serve tier above Build to compare against — a team that would have previously modeled a Build-to-Scale upgrade should instead request an Enterprise quote once volume clearly outgrows Build.
- Vapi's per-line concurrency fee ($10/month) pays for itself the moment it prevents even a few minutes of dropped or queued calls during a burst — and with a default of only 4 included lines (not 10), more teams will hit this threshold sooner than previously assumed.
- Retell's per-line concurrency fee ($8/month, beyond the first 20) is a smaller and less urgent cost lever for most teams, given the more generous free allowance.
Who pays more — and when the answer flips
For a lean, cost-optimized stack, Retell's componentized pricing — even though it isn't a single flat rate — consistently beats both Vapi's separately-billed pipeline and Bland's flat rate across all three volumes modeled, because its $0.055 voice-infrastructure charge plus a standard TTS voice already reflects an efficient default pairing that would cost more to assemble piece-by-piece on Vapi.
For a premium stack (top-tier LLM, premium TTS voice), Retell's own published ceiling ($0.31/min) and Vapi's premium pass-through estimate (~$0.198/min) both climb well above Bland's flat rate, at which point Bland becomes competitive or cheaper, particularly once a plan-tier fee is amortized — and Bland delivers that without requiring you to track, negotiate, or optimize three separate vendor bills. The practical ceiling on this comparison, though, is that Bland's own public pricing now stops at Build; above roughly 15,000–50,000 minutes/month, a real cost comparison against Bland requires a custom quote, not the numbers on its pricing page.
The practical decision isn't "which platform is cheapest" in the abstract - it's whether your use case needs the flexibility of choosing your own LLM/STT/TTS stack (Vapi, Retell) or whether Bland's narrower, bundled, predictable-rate model is a better operational fit, especially for teams without the engineering resources to monitor a multi-vendor cost stack in production.
For a managed OpenAI voice stack comparison, see GPT-Live-1 voice-agent costs and ElevenLabs Voice Agent production costs.
Limitations and uncertainty
- The lean and premium pass-through cost assumptions used for Vapi are illustrative composites based on commonly cited component rates (Deepgram, Twilio, and typical LLM/TTS pricing), not official Vapi rates for that specific bundle — your actual per-minute pass-through cost depends entirely on which providers and models you select.
- Retell's lean figure ($0.088/min) is built directly from Retell's own published component rates; its premium figure ($0.31/min) is Retell's own published range ceiling, not a specific bundle we priced out component-by-component, so your actual premium-configuration cost may land below $0.31/min depending on exactly which LLM and voice you choose.
- Vapi's and Retell's enterprise/custom pricing tiers were not modeled; both offer volume-based enterprise agreements that may undercut the self-serve rates used here at very high volume.
- Bland's Enterprise tier is custom-priced and was excluded from the 50,000-minute scenario for this reason; Bland does not publish a current rate for that tier, and figures referencing "Scale" should be treated as outdated because the tier no longer appears on its pricing page.
Official sources
- vapi.ai/pricing; docs.vapi.ai/billing/pricing-and-success-packages; docs.vapi.ai/billing/purchase-call-concurrency
- retellai.com/pricing (including its live component-pricing calculator)
- bland.ai/pricing (direct fetch, September 26, 2026)