Voice AI pricing is scattered across per-character TTS rates, per-minute voice-agent rates, and subscription tiers that don’t always mean the same thing between providers. Here’s a direct comparison using each provider’s own published pricing.
TTS cost per 1,000 characters
| Provider | Price per 1K characters |
|---|---|
| Lokutor | $0.005 |
| OpenAI | $0.015 - $0.03 |
| Deepgram | $0.03 |
| ElevenLabs | $0.05 - $0.10 |
Lokutor’s per-character TTS price is roughly 3-6x below OpenAI’s published range and roughly 10-20x below ElevenLabs’, depending on which ElevenLabs tier you’re comparing against. This isn’t a promotional rate - it’s a direct consequence of running TTS inference on commodity CPUs instead of GPU infrastructure, which is the largest recurring cost in most voice AI stacks.
Voice agent cost per minute
| Provider | Price per minute |
|---|---|
| Lokutor | $0.02 |
| Deepgram | $0.08 |
| ElevenLabs | $0.08 |
| OpenAI | n/a (no published per-minute voice-agent rate at time of writing) |
For a full conversational voice agent (not just TTS in isolation), Lokutor’s per-minute pricing runs about 4x below both Deepgram’s and ElevenLabs’ published per-minute rates.
Lokutor’s subscription tiers
Beyond raw per-unit pricing, Lokutor’s developer platform is available on monthly plans from $29 to $499/month with per-minute billing on top, Python and JavaScript SDKs, phone number provisioning, and usage analytics included. There’s no infrastructure to provision - you’re billed for usage, not for the CPU node running underneath it.
Why the gap exists
The short version: most competitors in this table run TTS and voice-agent inference on GPU-hosted autoregressive transformer models. GPU time is the dominant cost driver in that architecture, and it shows up directly in the price per character or per minute. Lokutor’s Versa TTS architecture uses flow matching instead of autoregressive generation, which runs efficiently on CPUs - removing the GPU cost entirely for that layer. We go into the architectural reasoning in more detail in our technical comparison of flow matching vs. autoregressive TTS.
A note on comparing prices fairly
Published pricing changes, and providers structure tiers differently (some bundle STT and LLM costs into a single per-minute rate, others price TTS and STT separately). The numbers above reflect each provider’s publicly listed pricing at time of writing, compared on the same units (per 1,000 characters, per minute) wherever both providers publish a comparable figure. If you’re evaluating providers for a specific workload, always confirm current pricing directly - and ask what’s actually included (concurrency limits, voice cloning, commercial usage rights) rather than comparing headline numbers alone.
Try it yourself
Lokutor’s developer platform includes free credits to test real pricing against your own workload - see our benchmarks page for the underlying latency and cost methodology, or get started directly.