ElevenLabs Alternatives in 2026, Priced

A cloned voice is the stickiest thing you can build on a text-to-speech API. The clone lives in the vendor's account, every character it speaks is metered, and each narrated lesson, video or support call makes the next migration a little harder.

On September 28, 2026, ElevenLabs shipped Eleven v4 at $0.08 per 1,000 characters over its API, while its web app sells the same characters as credits at roughly twice that rate. We priced one job, a US course platform narrating 1 million characters a month in its founder's cloned voice, on ElevenLabs and eight alternatives: $80 on Eleven v4, $49 on Cartesia, $15 on Fish Audio, and $10 on Murf, whose cloning is Enterprise-only.

How ElevenLabs Bills One Character Two Ways

ElevenLabs' API pricing, checked October 15, 2026, charges $0.08 per 1,000 characters for Eleven v4 and $0.04 for Eleven v4 Turbo, which the changelog puts at about 100 milliseconds of median inference latency. A plan fee is prepaid usage at that rate: Starter's $6 buys 75,000 characters of v4, Creator's $22 buys 275,000 and Pro's $99 buys 1,238,000.

The web app runs on credits instead. The same Creator plan includes 121,000 credits there, about 121 minutes, and extra Studio minutes cost $0.17 to $0.36 depending on the plan. The Pay As You Go docs confirm it: "API usage has lower rates than UI usage."

The free plan is non-commercial with attribution, so commercial rights and Instant Voice Cloning start at Starter. Professional Voice Cloning starts at Creator. Pay As You Go funds expire 12 months after purchase, and service pauses at a $0 balance unless Auto Top Up refills it. Eleven v4 also drops SSML and the Style and Speed sliders, so markup-driven pipelines need rework either way.

What 1 Million Characters in a Cloned Voice Costs

ElevenLabs' pricing page equates about 1,000 characters with a minute of audio, so the worked example is roughly 1,000 minutes, or 16.7 hours of narration a month, with a commercial license. The script prices that month on every service, plus a support line running 10,000 voice agent minutes.

js
Copy
// voice-bill.mjs - one month of narration in a cloned voice, priced on each service.
// usage: node voice-bill.mjs [characters] [agent minutes]
// Rates are from each vendor's own pricing page or docs, checked October 15, 2026.
const CHARS = Number(process.argv[2] ?? 1_000_000)
const MINUTES = Number(process.argv[3] ?? 10_000)
const perMillion = (rate, free = 0) => (Math.max(0, CHARS - free) / 1e6) * rate
const cheapest = (plans) => plans.reduce((a, b) => (b.cost < a.cost ? b : a))

// ElevenLabs API: plan fees are prepaid usage at the same per-1K rate. Starter ($6) is the
// cheapest plan with a commercial license and Instant Voice Cloning; Pay As You Go covers the rest.
const elevenApi = (perK, included) => 6 + (Math.max(0, CHARS - included) / 1000) * perK

// ElevenLabs Studio (web UI): credits, about 1,000 per minute, extra minutes priced per plan.
const studio = cheapest(
  [
    { plan: "Starter", fee: 6, credits: 30_000, extra: 0.2 },
    { plan: "Creator", fee: 22, credits: 121_000, extra: 0.18 },
    { plan: "Pro", fee: 99, credits: 600_000, extra: 0.17 },
    { plan: "Scale", fee: 299, credits: 1_800_000, extra: 0.17 },
  ].map((p) => ({ plan: p.plan, cost: p.fee + (Math.max(0, CHARS - p.credits) / 1000) * p.extra })),
)

// Cartesia: 1 credit = 1 character on Sonic-3.6; overage per 1M credits depends on the plan.
const cartesia = cheapest([
  { plan: "Pro", cost: 5 + perMillion(65, 100_000) },
  { plan: "Startup", cost: 49 + perMillion(45, 1_250_000) },
  { plan: "Scale", cost: 299 + perMillion(38, 8_000_000) },
])

const narration = [
  ["ElevenLabs v4 (API)", elevenApi(0.08, 75_000), "Starter + Pay As You Go, self-serve"],
  ["ElevenLabs v4 Turbo (API)", elevenApi(0.04, 150_000), "Starter + Pay As You Go, self-serve"],
  [`ElevenLabs Studio (${studio.plan})`, studio.cost, "web UI credits, self-serve"],
  [`Cartesia Sonic-3.6 (${cartesia.plan})`, cartesia.cost, "self-serve from the $5 Pro plan"],
  ["Fish Audio S2.1-Pro", perMillion(15), "self-serve over the API"],
  ["Murf Falcon 2", perMillion(10), "Enterprise plan only"],
  ["Deepgram Aura-2", perMillion(30), "no voice cloning"],
  ["Rime Mist v3", perMillion(30), "ordered enterprise clone"],
  ["Rime Coda", perMillion(50), "ordered enterprise clone"],
  ["Google Chirp 3 HD", perMillion(30, 1_000_000), "stock voices; first 1M free monthly"],
  ["Google Instant Custom Voice", perMillion(60), "allow-listed accounts only"],
].sort((a, b) => a[1] - b[1])

const usd = (n) => "$" + n.toLocaleString("en-US", { minimumFractionDigits: 2, maximumFractionDigits: 2 })
console.log(`${CHARS.toLocaleString("en-US")} characters of narration, one month\n`)
console.log("Service".padEnd(34) + "Cost".padStart(12) + "   Cloned voice")
for (const [name, cost, clone] of narration) console.log(name.padEnd(34) + usd(cost).padStart(12) + "   " + clone)
console.log("Chatterbox or Kokoro (self-host)".padEnd(34) + "$0 + compute".padStart(12) + "   Chatterbox only")

// Voice agents: platform rate per conversation minute, before your phone carrier.
const agents = [
  ["ElevenLabs Agents", 0.08, "+ LLM at cost"],
  ["Deepgram Voice Agent Standard", 0.075, "standard-tier LLM included"],
  ["Cartesia Managed Agents", 0.06, "+ $0.014/min on a Cartesia number"],
]
console.log(`\n${MINUTES.toLocaleString("en-US")} voice agent minutes, one month\n`)
for (const [name, rate, note] of agents) console.log(name.padEnd(34) + usd(rate * MINUTES).padStart(12) + "   " + note)

text
Copy
1,000,000 characters of narration, one month

Service                                   Cost   Cloned voice
Google Chirp 3 HD                        $0.00   stock voices; first 1M free monthly
Murf Falcon 2                           $10.00   Enterprise plan only
Fish Audio S2.1-Pro                     $15.00   self-serve over the API
Deepgram Aura-2                         $30.00   no voice cloning
Rime Mist v3                            $30.00   ordered enterprise clone
ElevenLabs v4 Turbo (API)               $40.00   Starter + Pay As You Go, self-serve
Cartesia Sonic-3.6 (Startup)            $49.00   self-serve from the $5 Pro plan
Rime Coda                               $50.00   ordered enterprise clone
Google Instant Custom Voice             $60.00   allow-listed accounts only
ElevenLabs v4 (API)                     $80.00   Starter + Pay As You Go, self-serve
ElevenLabs Studio (Pro)                $167.00   web UI credits, self-serve
Chatterbox or Kokoro (self-host)  $0 + compute   Chatterbox only

10,000 voice agent minutes, one month

ElevenLabs Agents                      $800.00   + LLM at cost
Deepgram Voice Agent Standard          $750.00   standard-tier LLM included
Cartesia Managed Agents                $600.00   + $0.014/min on a Cartesia number

Read the clone column before the cost column. Google, Murf, Deepgram and Rime use stock voices or put the clone behind a sales call, which makes Fish Audio at $15 the cheapest self-serve clone. Staying put has a cheap fix too: the same month costs $167 in the Studio and $80 over the API.

Cartesia: Credits That Map One to One to Characters

Cartesia sells Sonic-3.6, a 44-language model whose dated snapshot sonic-3.6-2026-08-27 never changes, at one credit per character. Its product page quotes first audio in under 90 milliseconds.

Cloning starts on the cheapest paid plan. Pro costs $5 a month and includes a commercial license plus Instant Voice Cloning from a 10 to 60 second clip. Startup adds two Professional Voice Clone slots trained on 30-plus minutes of audio, and Managed Agents cost $0.06 a minute.

The catch comes once you use up the plan. Overage runs $65 per million credits on Pro and $45 on Startup. Concurrency is tight as well: 3 requests on Pro, 5 on Startup.

Pricing: Pro $5/month for 100,000 credits · Startup $49 for 1.25 million · Scale $299 for 8 million · License: proprietary.

Fish Audio: Billed by the Byte, Cloning Included

Fish Audio charges $15 per million UTF-8 bytes for S2.1-Pro, an 83-language model, with no subscription or monthly minimum. Plain English text is one byte per character, so our month costs $15.

Creating the clone is one POST /model call with your sample, and the pricing page lists no charge for it. Until November 30, 2026, the same model also runs as s2.1-pro-free at $0 under fair-use limits.

That free variant promises nothing on time to first audio and comes with no DPA. Accounts are held to 5 concurrent requests until they have paid $100. Drama 3, launched October 13, 2026, had no rate on the pricing page.

Pricing: S2.1-Pro $15 per million UTF-8 bytes · License: proprietary API; the open S2 Pro weights are free for research and non-commercial use, and commercial use needs a Fish Audio license.

Murf Falcon 2: A Penny per 1,000 Characters

Murf sells Falcon 2 at 1 cent per 1,000 characters, the lowest hosted rate in our table, and reports a live 30-day median time to first audio of 95 milliseconds.

The rate is the draw, but you cannot clone a voice on your own. Murf's docs say Instant Voice Cloning "is available on the Enterprise plan only," and the endpoints return 403 until it is enabled.

Our month comes to $10. That buys 150-plus voices in 35 languages, served from us-east and us-west endpoints among others, with concurrency starting at 5 in US-East and 2 elsewhere. Murf also publishes a migration guide for Play.ai users, whose play.ht domain no longer resolved on October 15, 2026.

Pricing: Falcon 2 $0.01 per 1,000 characters · clones bill at the same rate · License: proprietary.

Deepgram: Agent Minutes With the LLM Included

Deepgram prices Aura-2 at $0.030 per 1,000 characters and Flux TTS, its newer agent-first model, at $0.045. Flux TTS reports exactly what a caller heard when they interrupt, and Deepgram quotes output starting in as low as 80 milliseconds.

The reason to pick Deepgram is a single bill for the whole agent. Its Voice Agent API costs $0.075 a minute on Pay As You Go, and that covers speech-to-text, a Standard-tier LLM such as gpt-5.4-mini, and the speech itself. New accounts get $200 in credit. Enterprise customers can self-host on NVIDIA GPUs.

There is no cloning, though. Deepgram's docs index has no voice cloning page, and Flux TTS voices speak English only.

Pricing: Aura-2 $0.030 per 1,000 characters · Flux TTS $0.045 · Voice Agent Standard $0.075/min · License: proprietary.

Rime: Fast Voices You Can Run On-Premises

Rime sells two models, Coda (released May 2026) and Mist v3 (March 2026), at $0.05 and $0.03 per 1,000 characters, with no plan fee on Starter.

Speed and placement are Rime's strengths. Its pricing page lists a P50 time to first audio of 37 milliseconds for Mist v3 and 96 for Coda. On Enterprise, both run on-premises or in your VPC via Docker Compose or Kubernetes, with a BAA for HIPAA.

Don't expect an instant clone. Each one is an ordered enterprise project that needs 30 to 60 minutes of clean recordings, and Starter caps you at 20 concurrent generations.

Pricing: Mist v3 $0.03 per 1,000 characters · Coda $0.05 · License: proprietary.

Google Cloud: A Free Million, Then Allow-Listed Clones

Google Cloud Text-to-Speech gives Chirp 3 HD voices the first 1 million characters free every month, then charges $30 per million.

With stock voices, our month costs nothing. A cloned voice is another matter. Instant Custom Voice costs $60 per million characters with no free tier, it is restricted to allow-listed accounts, and the speaker has to record a consent statement.

Pricing: Chirp 3 HD $30 per million characters after 1 million free · Instant Custom Voice $60 · License: proprietary.

Chatterbox: MIT-Licensed Cloning, Watermarked

Chatterbox is Resemble AI's open model family: Multilingual V3 (0.5B parameters, 23-plus languages), the English-only Turbo (350M) and Nano (110M), all cloning from a short reference clip.

Start with the trade-off: every file carries Resemble's PerTh watermark, and it survives MP3 compression.

The upside is commercial cloning at $0 per character. The README says Nano runs three times faster than real time on 8 CPU cores, so narration does not need a GPU.

License and stats: MIT · 26,955 GitHub stars as of October 15, 2026.

Kokoro: 82 Million Parameters, No Cloning

Kokoro is an 82-million-parameter model with Apache 2.0 weights, 54 voices in 8 languages, and 10,468,205 Hugging Face downloads in the month to October 15, 2026.

You get cheap narration on modest hardware under a permissive license, installed with pip install kokoro. It will not speak in your voice, because Kokoro has no cloning. Its latest package, 0.9.4, dates from April 5, 2025.

License and stats: Apache 2.0 · 9.3k GitHub stars on hexgrad/kokoro.

Side by Side

List prices checked October 15, 2026.

How to Choose Without Migrating Twice

1. Count last month's real characters. Export them from ElevenLabs' usage analytics and run node voice-bill.mjs <characters>; plan thresholds reorder the list at volume.
2. Keep the source recording, not the clone. Clones do not move between vendors. Cartesia re-clones from 10 to 60 seconds and Murf from up to 30, while Rime wants 30 to 60 minutes.
3. Render your three hardest scripts on two finalists, one with numbers, one with names and one long passage.
4. Read the weights license, not the repo's. F5-TTS code is MIT but its weights are CC BY-NC 4.0, and Fish's open S2 Pro weights are non-commercial too.
5. Time first audio from your own region, us-east-1 for most US stacks, because every latency number above is measured by the vendor.

Which One Should You Actually Use?

Narration in your own cloned voice on a budget: Fish Audio, $15 for our month with a clone created over the API.

A real-time agent in a cloned voice: Cartesia, with Instant Voice Cloning on the $5 plan and agents at $0.06 a minute.

An agent with the LLM on the same invoice: Deepgram's Voice Agent API at $0.075 a minute, if a stock voice is fine.

Healthcare or on-premises requirements: Rime Enterprise for a vendor with a BAA, or self-hosted Chatterbox when no audio may leave your network.

Staying on ElevenLabs: move Studio narration to the API, which cut our month from $167 to $80, and test v4 Turbo at $40.

Conclusion

ElevenLabs competes on quality with v4, while Murf, Fish Audio and Rime sell the same characters for $10 to $50 per million against v4's $80. OpenAI is retiring tts-1, tts-1-hd and both gpt-4o-mini-tts snapshots on January 6, 2027, in favor of its Realtime API. Before renewing, make sure that if this vendor raised prices tomorrow, you could re-clone your voice elsewhere from audio you still own.

Related DevToolLab Tools

• Character Counter - count the characters, spaces included, that per-character plans bill.
• Reading Time Calculator - estimate a script's speaking minutes.
• Invisible Character Remover - strip zero-width spaces that are billed and can trip a TTS engine.
• Speech to Text - play a clone's output into your browser's recognizer as an intelligibility check.

Related Guides

• Best Text-to-Speech APIs - blind-test rankings across more TTS vendors.
• Best AI Voice Agent Platforms - Vapi, Retell, Bland and ElevenLabs Agents per call minute.
• Best Speech-to-Text APIs - the recognition half of a voice agent.
• Best GPU Cloud Providers - where to host Chatterbox when CPU is too slow.