Last updated: September 2026 · By Shash Eran
ElevenLabs API Pricing for Developers 2026 — Cost Per Character, Quotas and Streaming Tiers
This guide is for developers calling ElevenLabs programmatically — what a character costs at each tier, how the quota and overage behave under a production workload, which tiers unlock WebSocket streaming and API voice cloning, and how the per-1M-character cost stacks up against OpenAI TTS, Google Cloud TTS and Amazon Polly. If you are buying a subscription to make voiceovers in the ElevenLabs dashboard rather than from code, the plan comparison, free tier limits and usage-based billing are covered on the ElevenLabs app pricing guide. Same character quotas underneath, different decision.
TL;DR
There is no separate API price list — API access is bundled into the same subscription tiers, so your effective cost is roughly $0.165–$0.22 per 1,000 characters depending on which tier you sit on. The free tier gives you an API key and enough characters to build and test an integration, but not to ship one. The two thresholds that actually decide your tier are architectural, not financial: WebSocket streaming starts at Creator (below it you are on request-response REST, which rules out real-time voice), and programmatic voice cloning starts at Starter and scales with your tier. ElevenLabs costs several times what OpenAI TTS or Google Cloud TTS charge per character — pick it when voice quality is a product differentiator, not when it is a checkbox.
September 2026 update
This guide was re-scoped to developer/API questions only — consumer plan comparison, free tier limits and usage-based billing now live on the separate app pricing guide. The characters-to-audio conversions in this page were also recalculated: the previous version over-stated audio hours per tier by roughly five times, and the corrected figures use ~900–1,000 characters per minute of speech, consistent with the per-minute reference below. The core tier lineup is unchanged: Free (10K chars), Starter ($5/mo, 30K chars), Creator ($22/mo, 100K chars), Pro ($99/mo, 500K chars), Scale ($330/mo, 2M chars), Business (custom, 10M+ chars) — treat those figures as last-verified rather than live, since ElevenLabs revises published pricing periodically. New additions in 2026: the Voice Design endpoint is now generally available on Creator and above — you describe a voice in natural language and ElevenLabs generates a custom voice without uploading samples. The recommended low-latency model is now eleven_flash_v2_5 (replaces Turbo v2); first-chunk latency over WebSocket is approximately 150–180ms under standard conditions. ElevenLabs also introduced an Audio Isolation endpoint (background noise removal) available on Pro and above at no additional character cost. The OpenAI TTS comparison row in this guide has been updated to reflect the current gpt-4o-mini-tts model.
1. What API Access Costs at Each Tier
The first thing to understand is that ElevenLabs does not sell API access as a separate product. There is no pay-as-you-go developer SKU with its own rate card. You subscribe to a tier, that tier carries a monthly character quota, and your API calls draw down the same quota that dashboard generations do. An API key is issued on every tier including Free.
That has a practical consequence worth planning around: your API capacity and your team's manual usage share one bucket. If someone spends an afternoon auditioning voices in the dashboard, your production integration has less headroom that month. On anything past a prototype, put the API integration on its own workspace or account so a human cannot exhaust a service quota by accident. The table below shows what each tier gives a developer — quota, cloning capacity, and the transport that decides whether real-time is possible at all.
| Plan | Price/mo | Characters | Custom Voices | Streaming |
|---|---|---|---|---|
| Free | $0 | 10,000 | 0 (no cloning) | REST only |
| Starter | $5 | 30,000 | 1 (IVC only) | REST only |
| Creator | $22 | 100,000 | 10 (IVC + PVC) | WebSocket ✓ |
| Pro ⭐ | $99 | 500,000 | 30 (IVC + PVC) | WebSocket ✓ |
| Scale | $330 | 2,000,000 | 160 (IVC + PVC) | WebSocket ✓ |
| Business | Custom | 10M+ | Custom | WebSocket ✓ |
IVC = Instant Voice Cloning (fast, from short samples). PVC = Professional Voice Cloning (high-fidelity, requires longer samples, processing time 1–24 hours).
Is the ElevenLabs API free?
Partly, and the distinction matters if you are scoping a build. Yes, you get a real API key on the free tier — no credit card, no sales call. You can generate a key in the dashboard, hit the text-to-speech endpoint, stream a response, and confirm your audio pipeline works end to end. For evaluating whether ElevenLabs is the right engine, the free tier is entirely sufficient and you should use it before paying anything.
No, it is not free in a way you can ship on. Three constraints bite the moment you go past a prototype:
- Volume. The free character allowance is roughly ten minutes of audio a month. A single integration test suite that synthesises fixtures will burn through it. Budget on the assumption that free covers your first afternoon, not your first sprint.
- Licence. Free-tier output is non-commercial with attribution expected. Anything user-facing in a product you charge for needs a paid tier, regardless of how few characters you use.
- Transport. Free and Starter are REST-only. If your architecture depends on WebSocket streaming — any conversational or real-time voice feature — the free tier cannot even demonstrate the shape of the thing you are building. Prototype real-time on Creator or you will design against the wrong latency profile.
The pragmatic path: build and validate on Free, move to Starter ($5) the moment anything is user-facing and non-realtime, and jump to Creator when you need streaming or programmatic cloning. Do not linger on Free hoping to prove out a production design — the rate limits and REST-only transport will give you misleading performance numbers.
2. How Character Quotas Work
ElevenLabs measures API usage in characters — every character you send to the text-to-speech endpoint counts against your monthly quota, including spaces and punctuation. Understanding this helps you accurately estimate your tier requirements.
Character Quota Reference
- Average word: ~5 characters
- 1 minute of speech: ~800–1,000 characters
- 10-minute audio: ~8,000–10,000 characters
- 30-minute podcast episode: ~24,000–30,000 characters
- Full audiobook chapter (~5K words): ~25,000 characters
- Free (10K): ~10 minutes of audio/mo
- Starter (30K): ~30 minutes of audio/mo
- Creator (100K): ~1 hr 45 min of audio/mo
- Pro (500K): ~8–9 hours of audio/mo
- Scale (2M): ~33–37 hours of audio/mo
Characters do not roll over between billing periods. If you consistently use less than your quota, you should consider downgrading — unused characters are money left on the table. If you're regularly hitting overages, upgrading to the next tier almost always has a lower cost-per-character than paying overage rates.
3. Cost-Per-Character Math at Each Tier
To compare tiers on efficiency, here's the effective cost per 1,000 characters at each tier's base quota. The audio-hour column converts at ~54,000 characters per finished hour (roughly 900 characters per minute of speech):
| Plan | Price/mo | Characters | $/1K chars | $/hour audio |
|---|---|---|---|---|
| Starter | $5 | 30K | $0.167 | ~$9.00 |
| Creator | $22 | 100K | $0.220 | ~$11.90 |
| Pro | $99 | 500K | $0.198 | ~$10.70 |
| Scale | $330 | 2M | $0.165 | ~$8.90 |
The Creator tier appears most expensive per character on paper, but it's the entry point for WebSocket streaming and Professional Voice Cloning — features that are essential for production applications. The Pro tier hits the best balance of per-character efficiency, feature unlocks, and quota depth for most serious developers and production apps.
4. WebSocket Streaming — Which Tiers Support It and Why It Matters
WebSocket streaming lets your application receive audio data as it's generated — character by character, in real time — rather than waiting for the entire audio file to be synthesized before playback begins. This is the difference between a conversational AI that responds with sub-second audio latency and one that makes users wait 3–5 seconds per response.
WebSocket streaming is available from Creator tier ($22/mo) and above. The Free and Starter tiers only support REST API, which uses a request-response model — you send text, wait for synthesis, receive the audio file. REST is fine for batch audio generation (podcast production, audiobook creation) but is not suitable for real-time conversational applications.
- Voice chatbots and conversational AI assistants
- Real-time phone call voice AI (IVR, sales automation)
- Live streaming with AI voice narration
- Interactive games and virtual characters that respond to player input
- Accessibility tools that read content aloud as it's generated
- Batch podcast production (synthesize full scripts offline)
- Audiobook creation from manuscript text
- E-learning content narration (pre-recorded)
- Marketing video voiceovers
- Content localization (translating and voicing articles)
5. Voice Cloning Tiers and Limits
ElevenLabs offers two types of voice cloning: Instant Voice Cloning (IVC) and Professional Voice Cloning (PVC). Understanding the difference — and which tier you need — is essential if custom voices are part of your use case.
- Create a voice from as little as 1 minute of audio
- Processing is near-instant (seconds)
- Quality: good but perceptibly different from the source speaker at close inspection
- Use case: quick prototyping, consistent character voices, internal tools
- Available from: Starter tier (1 voice)
- Requires 30+ minutes of clean, high-quality audio samples
- Processing takes 1–24 hours
- Quality: near-indistinguishable from original speaker in many cases
- Use case: brand voice, celebrity partnerships, publisher content monetization, executive communications
- Available from: Creator tier ($22/mo) and above
| Plan | Custom Voices | Cloning Type |
|---|---|---|
| Free | 0 | No cloning |
| Starter | 1 | IVC only |
| Creator | 10 | IVC + PVC |
| Pro | 30 | IVC + PVC |
| Scale | 160 | IVC + PVC |
| Business | Custom | IVC + PVC + Custom Models |
6. ElevenLabs vs OpenAI TTS vs Google Cloud TTS vs Amazon Polly
ElevenLabs is not the cheapest TTS API available — it's the highest quality. Here's an honest comparison against the main alternatives at equivalent character volumes:
| Provider | Cost/1M chars | Voice Quality | Voice Cloning | Streaming |
|---|---|---|---|---|
| ElevenLabs Pro | ~$198 | Best-in-class | Yes (IVC + PVC) | WebSocket |
| OpenAI TTS (gpt-4o-mini-tts) | ~$60 | Good, natural | Limited | Streaming |
| Google Cloud TTS Neural2 | ~$16 | Good, robotic edge | No | Streaming |
| Amazon Polly Neural | ~$16 | Decent, flat delivery | No | Streaming |
When to choose ElevenLabs over cheaper alternatives:
- You need emotionally expressive voice — joy, urgency, sadness — not just neutral reading
- You need voice cloning (custom speaker voices)
- Your use case involves long-form listening (audiobooks, podcasts) where listener fatigue from robotic voices is a real concern
- You need high-quality multilingual support — ElevenLabs supports 29+ languages with genuine naturalness
- The end product's quality directly affects your revenue (premium content, brand voice)
When Google TTS or Amazon Polly is sufficient: Internal tools, notifications, accessibility features where cost matters more than quality, high-volume utility applications where listeners won't spend more than 30 seconds listening at a time.
7. Which Tier Is Right for You?
10K characters is roughly 10 minutes of audio. That is enough to evaluate voice quality and wire up your API integration end to end, but nowhere near a production use case — you will exhaust it during a single afternoon of testing. The Free tier also has rate limits and no voice cloning.
30K chars/mo is ~30 minutes of audio. Good for a solo developer building personal tools, small-scale content production (1–2 short podcast episodes/mo), or experimentation. No streaming support limits production app use.
100K chars/mo (~1 hr 45 min of audio), WebSocket streaming, 10 custom voices, PVC access. This is the minimum viable tier for production voice AI applications. Real-time chatbots, small SaaS products, early-stage startups.
500K chars/mo (~8–9 hours of audio), 30 custom voices, all streaming features. This is the sweet spot for most production applications — enough volume to run a real product, sufficient voice slots for most use cases, and the per-character rate is better than Creator.
2M chars/mo (~33–37 hours of audio), 160 custom voices. For audio-first products with significant user bases, content publishers generating large volumes of spoken content, or platforms offering voice features as a core product component.
10M+ chars/mo, custom voice limits, dedicated SLA, priority support, and potentially custom model training. Contact ElevenLabs sales directly for pricing. For media companies, enterprise software, and platforms where voice is mission-critical infrastructure.
Start Building with ElevenLabs API
Free tier available — a real API key and enough characters to wire up and test your integration. Upgrade when you're ready to go live.
WebSocket streaming, voice cloning, 29+ languages — the highest-quality TTS API available.
Try ElevenLabs Free →Frequently asked questions
How much does the ElevenLabs API cost?
ElevenLabs API pricing tiers: Free ($0, 10K chars/mo — test only), Starter ($5/mo, 30K chars), Creator ($22/mo, 100K chars), Pro ($99/mo, 500K chars), Scale ($330/mo, 2M chars), Business (custom, 10M+ chars). All paid tiers include the full voice library and REST API. WebSocket streaming starts at Creator tier.
Is the ElevenLabs API free?
You get a real API key on the free tier with no credit card, so it is free to build and test against — but not to ship on. The free character allowance is roughly ten minutes of audio a month, free-tier output is non-commercial with attribution expected, and Free and Starter are REST-only, so you cannot accurately prototype a real-time or conversational feature below the Creator tier where WebSocket streaming begins.
Is ElevenLabs API cheaper than OpenAI TTS?
No — ElevenLabs is more expensive per character than OpenAI TTS (gpt-4o-mini-tts at ~$60/1M chars vs ElevenLabs Pro at ~$198/1M chars). However, ElevenLabs delivers significantly higher voice quality, emotional range, and voice cloning capabilities that OpenAI TTS doesn't match. For applications where voice quality is a core product differentiator, the premium is usually justified.
Does ElevenLabs API support WebSocket streaming?
Yes — WebSocket streaming is available from Creator tier ($22/mo) and above. Streaming allows your application to receive audio data as it's generated in real time, enabling sub-second latency for conversational AI, voice chatbots, and interactive applications. Free and Starter tiers support REST API only (request-response), which has higher latency and is not suitable for real-time applications.
How many custom voices can I clone with the ElevenLabs API?
Voice cloning limits by tier: Free (0 custom voices), Starter (1 voice, IVC only), Creator (10 voices, IVC + PVC), Pro (30 voices, IVC + PVC), Scale (160 voices, IVC + PVC), Business (custom). Instant Voice Cloning creates a voice from ~1 minute of audio in seconds. Professional Voice Cloning requires 30+ minutes of clean samples and produces near-indistinguishable voice replicas with 1–24 hours of processing time.
What is ElevenLabs' character quota and how does it work?
ElevenLabs measures usage in characters — every character you send to the API (including spaces and punctuation) counts toward your monthly quota. A typical spoken word averages ~6 characters including the space. 100K characters is roughly 16,000–17,000 spoken words (~1 hour 45 minutes of finished audio). Unused characters do not roll over at the end of each billing cycle. Exceeding your quota triggers per-character overage charges, which are more expensive than the bundled rate — upgrading to the next tier is usually more cost-effective than paying overages.
Written by Shash
Founder of Infinfy Solutions · Real tools, real client work.