ElevenLabs pricing in 2026: plans, credits, and agent costs

The AI agency PxlPeak published its own ElevenLabs invoice: $303 for January 2026, covering voice content for six clients, IVR systems for two, and live voice agents for three. What surprises their clients isn’t the total. It’s the mix. As founder John V. Akgul put it, “conversation minutes dwarf TTS consumption inside a month.”

That’s the ElevenLabs story in one invoice. Most buyers arrive pricing a voiceover tool and end up running a metered, usage-billed voice platform inside their product. If that’s already you, and the bill is harder to read than the transcripts, a CloudZero demo will show you voice spend allocated per feature and per customer. If you’re still comparing plans, read on. Both readers get their answer here.

How much does ElevenLabs cost in 2026?

ElevenLabs costs $0, $6, $22, $99, $299, or $990 per month depending on tier, with Enterprise priced by quote. Every plan meters usage through a monthly pool of credits, and the same six price points apply across ElevenLabs’ three product families: ElevenCreative, ElevenAgents, and ElevenAPI.

Here is the current lineup:

A caution before you cross-check these numbers elsewhere: ElevenLabs cut prices during 2026, and plenty of ranking guides still show the old rate card, including Scale at $330. The figures here come from the live pricing page, dated at the top of this article, and pricing this fluid deserves a fresh look on the day you commit a budget.

One structural thing trips up almost every first-time buyer: those six prices apply three times over. ElevenLabs splits its pricing into ElevenCreative for voiceover, dubbing, music, and content production, ElevenAgents for conversational voice agents, and ElevenAPI for developers embedding speech into products.

Each family shows the same dollar tiers but meters different things: credits for creative work, call minutes for agents, characters for the API. Knowing which family you’re actually buying is the first budgeting decision, and it decides which of the tables here applies to you. Many teams end up on two families at once, creative credits for marketing and agent minutes for support, which is two meters before the LLM meter even enters.

The feature gates matter as much as the prices. Commercial use starts at Starter, professional voice cloning starts at Creator, API access to 44.1kHz PCM audio starts at Pro, and the famous “low-latency TTS as low as 5 cents per minute” line is a Business-tier feature. Free-plan audio carries no commercial rights and must credit ElevenLabs when published, which makes it a sandbox, not a production tier.

How do ElevenLabs credits actually work?

One credit equals roughly one character of text-to-speech on the standard Multilingual v2 model, and newer models draw 0.5 to 1 credit per character. About 1,000 characters produces a minute of natural-paced narration, so credit pools translate to the approximate minute counts shown above. Treat the conversion as a planning estimate: reading speed, punctuation, and model choice all move it.

Credits reset monthly and unused credits don’t roll over unless you upgrade plans, which quietly penalizes lumpy production schedules. A documentary team that records everything in one week each quarter burns overage in the busy month and donates unused credits back in the quiet ones. Usage-billed services reward steady consumption, and voice is no exception.

Once you pass your included pool, two overage meters exist. In the app, extra minutes bill at your plan’s rate, the extra-minute column above, which falls from roughly $0.36 on Free to $0.17 on Pro and up. Through the API, overage bills per model instead: Flash runs $0.05 per 1,000 characters and Multilingual $0.10, a clean 2x gap that makes model selection your first cost decision. Other products draw the pool at their own rates: transcription is far cheaper per minute than synthesis, while dubbing consumes credits fastest of all.

The plan-versus-overage math has a known crossover. When your overage spend regularly reaches 30 to 50% of the next tier’s price, upgrading beats staying put. Annual billing takes roughly 17% off everything, the equivalent of two free months, which is the easiest discount in this article to claim and the most commonly skipped.

How does the ElevenLabs API bill for developers?

The ElevenLabs API runs on the same six subscription prices, but it has its own rate card: ElevenLabs’ API pricing page publishes per-model rates and per-model monthly inclusions.

Text-to-speech bills at $0.05 per 1,000 characters on Flash and v3 Conversational, and $0.10 per 1,000 on v3 and Multilingual v2. Each plan’s included characters depend on the model you pick: Scale includes 2.99 million Multilingual characters a month, or 5.98 million on Flash.

Plan choice still matters for developers, because capability gates ride the tiers. API access to 44.1kHz PCM audio starts at Pro, and concurrent request limits climb from 2 on Free to 25 on Business. A product that generates audio for many users simultaneously buys tiers for the same reason a call center does: throughput, not price breaks.

Worked example for an app that narrates articles, generating 2 million characters per month. On Multilingual, the Pro plan includes 990,000 characters for $99, leaving just over 1 million characters of overage at $0.10 per 1,000: about $101, for a monthly total near $200. Scale at $299 absorbs the full 2 million inside its 2.99 million included characters. The same volume on a Creator plan means roughly 1.78 million Multilingual overage characters, which is how a $22 subscription becomes a $200 invoice.

That gap is the recurring theme of usage-based AI: the plan price is a floor, and product success raises the ceiling. Every new user of your voice feature spends your credits, which is a wonderful problem exactly as long as someone is watching the meter per feature rather than discovering it per invoice.

What does ElevenAgents cost per minute?

Voice agents cost $0.08 per additional minute on every single tier, from Free to Business. What changes as you climb tiers isn’t the minute rate. It’s how many minutes come included and how many calls you can run at once.

Run the division and something clarifying appears: Pro’s included minutes work out to exactly 8 cents each, and so do Scale’s and Business’s. The plans are prepaid minutes at the standard rate. What you’re actually buying with a bigger plan is concurrency, the right to answer 40 phone calls at once instead of 4. For a support line, concurrency is capacity planning, and capacity is what the tier ladder sells.

Two line items surround that 8 cents. Text messages bill at $0.003 each, cheap enough to ignore until an agent starts confirming every appointment by SMS at volume. And burst pricing charges $0.16 per minute, double rate, for calls that exceed your concurrency cap.

Burst is a feature, not a penalty: it keeps your line answering during a spike. But a marketing campaign that doubles call volume on a Creator plan pays 2x on the overflow, which is the kind of surprise that deserves a forecast, not a discovery.

What’s not in the 8 cents? The three meters on every call

The $0.08 per minute covers only the ElevenLabs layer. Every production voice agent call runs three meters at once: ElevenLabs minutes, LLM tokens for the model doing the thinking, and telephony for the phone line itself. ElevenLabs bills LLM usage separately based on the model you choose and passes telephony through at cost.

That architecture has a direct finance consequence: one customer phone call generates charges across up to three vendors, on separate invoices, in different units. Your agent’s brain might run on OpenAI, Claude, or Gemini, each with its own token meter. The phone leg bills through your telephony provider. No line on any of those three invoices says “cost per call.”

Sketch a four-minute support call to see the shape of it. The ElevenLabs layer costs $0.32. The LLM leg depends entirely on model choice and conversation length: a budget model like DeepSeek might add a cent or two, a frontier model with a long context considerably more. Telephony adds its per-minute carrier rate on top.

The precise total varies by stack, and that variance is the point. The advertised 8 cents is the only number anyone quotes, and it’s one meter of three.

This is the voice version of a pattern we’ve measured across AI workloads, where the true cost of a resolved AI task is often 10 to 50 times the posted per-call price once the full stack is counted. Longer conversations compound all three meters simultaneously.

A chatty customer doesn’t just cost more ElevenLabs minutes; every extra exchange grows the LLM context, and inference costs scale with context, so late-call tokens cost more than early ones. Call length is set by the customer, not by you, which makes it the one cost driver in your voice stack that you can influence but never control.

What does ElevenLabs cost at scale?

At production volume, the ElevenLabs layer alone reaches four figures monthly, and the full voice stack typically lands well above it. A support operation handling 20,000 agent minutes per month pays $990 for the Business plan, then $610 for the 7,625 minutes beyond its included 12,375, or $1,600 before a single LLM token or phone-line minute is counted.

The same spend reads very differently depending on what those minutes accomplish. Take 20,000 minutes resolving 5,000 calls at different resolution rates, counting the ElevenLabs layer alone:

Identical invoice, a 3.6x swing in unit cost, and the invoice itself can’t tell you which row you’re in.

Whether the all-in number is good news depends on unit economics no invoice will hand you. If those 20,000 minutes resolved 5,000 support calls that would have cost $6 each with human agents, the voice stack is printing money. If half those minutes were an agent cheerfully failing to resolve anything before escalating to a human anyway, you’re paying twice per contact and the topline spend hides it.

The data to answer it lives in three places. The ElevenLabs charge knows minutes, the LLM bill knows tokens, and neither knows which customer called or whether the issue got resolved.

This three-invoice puzzle is CloudZero’s home turf. It ingests spend from AI platforms like OpenAI and Anthropic alongside your cloud bills, pulls metered vendors like ElevenLabs in through AnyCost, and allocates everything to the units that matter: cost per call, per resolution, per customer.

Concretely, a voice team running ElevenLabs through CloudZero sees things no single invoice contains. The true cost of the support agent feature, all three meters combined, trending week over week. Which customers drive the longest, most expensive calls. Whether last month’s prompt changes actually shortened conversations or just moved cost between meters. And an alert when agent minutes spike, instead of a surprise thirty days later.

There’s a durability angle too. ElevenLabs repriced during 2026, and it won’t be the last time. Teams that know their per-minute exposure by feature can re-forecast a rate change over lunch. Teams that know one blended monthly total start that conversation from zero, every time.

The LLM leg gets even more visible if your agents route model calls through a gateway. CloudZero’s AI Signals, currently in preview, ingests through LiteLLM and Bifrost gateways, OpenTelemetry, or a macOS collector, and surfaces live usage in seconds, with costs reconciled against provider billing within about 24 hours. That turns a runaway agent loop from a month-end mystery into a same-day alert.

Duolingo, Skyscanner, and Coinbase run their cloud and AI spend through CloudZero for exactly this kind of unit-level visibility. If voice is becoming a real line item for you, take the self-guided tour and see how per-call economics look when all three meters land in one view.

How do you keep ElevenLabs costs under control?

Five levers move the ElevenLabs bill, and the biggest ones live in your agent design, not your plan selection. Teams that manage voice spend well choose models deliberately, engineer calls shorter, cap concurrency intentionally, claim the boring discounts, and measure cost per resolution instead of cost per month.

Voice is following the path text AI walked two years ago: from novelty budget to real COGS. The ElevenLabs meter compounds with usage, the LLM meter compounds with conversation length, and the telephony meter runs the whole time.

The companies that come out ahead won’t be the ones that spent the least on ElevenLabs. They’ll be the ones who knew what every minute bought. Take the self-guided tour and see how per-call economics look with all three meters in one view.

Frequently asked questions about ElevenLabs pricing

Author Spotlight

Lyne Carolyne

Lyne Carolyne has several years of experience in AI and cloud economics and brings that understanding into the content she creates. Outside work, she's an avid explorer.