Cartesia's pricing is refreshingly legible: 1 credit = 1 character of synthesized speech. There's a free tier, a cheap Pro plan around $5/month, a Scale plan at $299/month for 8 million credits, and custom Enterprise. The catch isn't complexity — it's that per-character billing quietly adds up once a voice agent is talking to real users all day.
The quick answer
Start free with 20,000 credits. Pro is about $5/month ($4 on annual billing) and adds unlimited instant voice cloning and commercial use. Scale is $299/month for 8,000,000 credits plus higher concurrency. Effective cost lands around $5–$37 per million characters depending on tier — cheaper than ElevenLabs, pricier than budget TTS like Gemini Flash.
Every plan compared
| Plan | Price | Credits | Best for |
|---|---|---|---|
| Free | $0 | 20,000 | Prototyping, testing latency and voices before you commit. |
| Pro | ~$5/mo ($4 annual) | Higher monthly allotment | Solo builders and small apps; unlocks unlimited instant cloning + commercial use. |
| Scale | $299/mo | 8,000,000 | Production voice agents with steady volume and concurrency needs. |
| Enterprise | Custom | Negotiated | High-volume deployments needing volume rates, SLAs and dedicated capacity. |
New to the platform? Read the full Cartesia review first — the speed and cloning story is what justifies the price.
What a credit buys
Because a credit equals a character, it's easy to translate into real output. Roughly 1,000 characters is a short paragraph or about 10–15 seconds of speech. So the free tier's 20,000 credits is only a few minutes of audio — fine for testing, small for production. Scale's 8 million credits is millions of characters, which covers a lot of narration or many minutes of agent conversation each month.
The variable that trips people up is how much your agent talks. A chatty assistant that reads long responses burns characters fast; a terse one sips them. Estimate average response length × call volume, not just call count.
Cost at scale
At the per-million-character level, Cartesia is mid-market: $5–$37 per million depending on plan. That undercuts ElevenLabs on most tiers, which is a big reason it shows up in voice-agent stacks. But it's more expensive than the cheapest models, so if you're generating huge volumes of low-stakes speech, run the numbers against a budget TTS too.
One more cost note for voice agents: Cartesia is usually just the TTS layer. In a full stack you also pay for speech-to-text, the LLM and telephony — often routed through an orchestrator like Vapi, whose own pricing stacks on top. Budget the whole pipeline, not just the voice.
Is Cartesia worth it?
If you're building real-time voice and latency is the priority, yes — Cartesia's speed and 3-second cloning are worth paying for, and the per-character model is transparent enough to forecast. Start free, prototype, then size Pro or Scale to your measured usage.
If your needs are occasional narration, broad multilingual coverage, or a big ready-made voice catalog, compare against ElevenLabs and see our best AI voice generators roundup before deciding — a different tool may be cheaper for your specific mix.
Frequently Asked Questions
How much does Cartesia cost in 2026?
Cartesia has a free tier with 20,000 credits, a Pro plan around $5/month ($4 on annual billing), Scale at $299/month for 8 million credits, and custom Enterprise. It charges 1 credit per character of synthesized text.
What is a Cartesia credit worth?
One credit equals one character of text-to-speech. So 8 million credits on the Scale plan is roughly 8 million characters — enough for a lot of narration or many minutes of voice-agent conversation, depending on how much your agent speaks.
What does Cartesia cost per million characters?
Effective pricing works out to roughly $5–$37 per million characters depending on your plan and volume — cheaper than ElevenLabs, but more expensive than budget options like Gemini Flash TTS. Higher tiers lower the per-character rate.
Is Cartesia's free plan enough to build with?
The 20,000-credit free tier is generous for prototyping and testing latency and voices, but it's small for production — 20,000 characters is only a few minutes of speech. You'll move to Pro or Scale once you go live.
Is Cartesia worth it?
For real-time voice agents where latency is critical, yes — Cartesia's speed and 3-second cloning justify the cost. For occasional narration or broad multilingual needs, a per-character model can get pricey, and ElevenLabs or a budget TTS may fit better.
Do I pay for more than TTS in a voice agent?
Usually, yes. Cartesia is typically just the text-to-speech layer. A full voice agent also incurs speech-to-text, LLM and telephony costs, often via an orchestrator like Vapi, so budget the whole pipeline rather than voice alone.