ElevenLabs
Listed at https://elevenlabs.io
💰 Token Pricing
| Type | Price | Note |
|---|---|---|
| Input | TTS: $0.03-0.30/1K chars depending on model/quality; STT: $0.47/hour; Sound Effects: $0.04/request | per million tokens |
| Output | Same as input rate per model | per million tokens |
🤖 Supported Models (8)
✨ Pros
- ✓Industry-leading TTS quality across 29+ languages
- ✓Voice Cloning (Professional + Instant): accurate voice replication
- ✓ElevenAgents: conversational AI voice assistant API
- ✓ElevenScribe: high-accuracy speech-to-text at $0.47/hour
- ✓ElevenSound: AI sound effects generation at $0.04/request
- ✓44.1kHz PCM high-fidelity audio output (Pro plan)
- ✓Startup Grants Program (12 months free, 33 million characters)
⚠️ Cons
- ×Charged by characters/hours, not tokens — cost comparison with LLMs is not straightforward
- ×China access requires stable proxy
- ×Free tier is limited (10,000 credits/month ≈ 10 minutes audio)
- ×Voice quality varies by language (English strongest, smaller languages weaker)
- ×Not an LLM API — not suitable for text generation/reasoning tasks
🎯 Best For
High-quality TTS applications (podcasts, audiobooks, voice assistants); voice cloning and customization; voice feature integration in SaaS products
💰 Pricing & Plans
| Plan | Price | Credits / Quota | Best For |
|---|---|---|---|
| Free | $0 / month | 10K credits/month (~10 min audio); 3 Studio projects; non-commercial use | Trying ElevenLabs voices before upgrading |
| Starter | $6 / month | 30K credits/month; 20 Studio projects; commercial license; Instant Voice Cloning | Indie creators publishing short audio |
| Creator (Popular) | $22 / month ($11 first month) | 100K credits/month; Professional Voice Cloning; Music commercial use; Dubbing Studio | Podcasters and audiobooks at hobby-professional scale |
| Pro | $99 / month | 500K credits/month; 44.1 kHz PCM via API; 192 kbps audio; 3 Professional Voice Clones | Production SaaS integrating TTS at moderate volume |
| Scale | $299 / month | 1.8M credits/month; 3 workspace seats; 3 Professional Voice Clones; team collaboration | Mid-size teams running multiple voice apps |
| Business | $990 / month | 6M credits/month; 10 seats; low-latency TTS as low as 5¢/min; 10 Professional Voice Clones | High-volume call centers and real-time voice agents |
| Enterprise | Custom quote | Custom credits/seats; SSO/SAML; HIPAA BAA; data residency; priority support; elevated concurrency limits | Regulated industries and Fortune-500 deployments |
| Startup Grants Program | $0 (12 months) | 33M characters valid for 12 months; full API access for builders and early-stage startups | Early-stage startups launching a voice-first product |
🔧 API & Developer Experience
- •Protocol: REST + WebSocket dual transport — REST for batch TTS/STT generation, WebSocket for low-latency conversational streaming (sub-200ms first-byte on Business tier)
- •Official SDKs: Python (`pip install elevenlabs`) and Node.js (`npm install @elevenlabs/elevenlabs-js`) — both maintained by ElevenLabs; community SDKs in Go, Ruby, Rust, .NET, Java
- •Audio formats & streaming: Output: MP3, PCM (16/24/32-bit), WAV, μ-law; 44.1 kHz PCM available on Pro and above; streaming chunked responses supported on TTS endpoint for real-time playback
- •Voice cloning APIs: Instant Voice Cloning (≤1 min sample, available Starter+) and Professional Voice Cloning (≥30 min sample, Creator+) via `/v1/voices/add`; consent-verification metadata required
- •ElevenAgents / Conversational: Conversational AI voice agents via WebSocket — turn-taking, interruption handling, tool calling; ElevenAgents Platform endpoint combines STT + LLM + TTS in one pipeline
- •MCP server: Hosted remote MCP server at `api.elevenlabs.io/v1/mcp` (global) and `api.us.elevenlabs.io/v1/mcp` (US residency) — OAuth-authenticated, exposes agents and creative generation as tools to any MCP client
- •Billing model: Unified credit pool across all products (TTS, STT, Music, Sound Effects, Dubbing, Voice Cloning); credits roll over up to 2× monthly quota; usage-based top-ups above plan quota on Creator+
🎙️ Voice Quality & Multilingual Capability
ElevenLabs remains the industry benchmark for natural-sounding neural TTS, and that edge is most visible on multilingual output. Eleven Multilingual v2 covers 29 languages with full emotional prosody and code-switching, while Eleven Turbo v2.5 and Eleven Flash v2 trade a small amount of expressiveness for ~50% lower latency and credit cost, making them the default choice for real-time conversational agents. Beyond text-to-speech, the platform bundles speech-to-text (ElevenScribe at $0.47/hour with speaker diarization and character-level timestamps), ElevenSound effects generation, Eleven Music for background tracks, and an automatic Dubbing Studio that translates and re-voices long-form audio across 70+ languages. The voice-cloning APIs are the second pillar: Instant Cloning needs roughly one minute of clean audio, while Professional Cloning produces studio-grade replicas from 30+ minutes of recorded speech with consent-verification metadata built into the API surface. ElevenAgents ties these primitives together into a hosted conversational-AI pipeline, which is why ElevenLabs has become the default voice layer for production chatbots and AI call centers in 2026.
🌐 China Access & Latency
ElevenLabs is not deployed inside mainland China and has no announced mainland-availability roadmap, so calling api.elevenlabs.io from a Chinese IP carries the familiar cross-border tax — high first-packet latency (typically 300-800 ms to the closest overseas endpoint in Singapore or Tokyo), occasional timeouts during peak hours, and the risk that the workspace console will flag sign-ins from unfamiliar geographies. Production users in China typically route through a stable overseas proxy or a Hong Kong relay, and the Business-tier low-latency TTS feature (as low as 5¢/minute) is only available once connected through that relay. The graceful degradation path is the API-only posture: the HTTP and WebSocket endpoints are reproducible, so a self-hosted proxy or VPC peering into ElevenLabs' us-east region works for batch generation, but ElevenAgents' conversational pipeline adds extra round-trips and is most sensitive to China network conditions. For teams needing full onshore synthesis, pairing an open-weight model like CosyVoice or a domestic TTS SaaS (iFlytek, Aliyun) for in-China traffic and reserving ElevenLabs for overseas quality-critical use is the common 2026 split.