DEV Community

Daniel Varela
Daniel Varela

Posted on

ElevenLabs alternative for real-time voice agents: CPU-native, 2 cents a minute

ElevenLabs is the default answer for synthetic voices, and for good reason: a large voice library, strong voice cloning, dubbing, and a broad creative suite. If you are producing narration, dubbing video, or exploring voices, it is the right tool.

If what you are building is a real-time voice agent, the requirements change. You need streaming time to first audio that survives a phone call, a per-minute price that holds up at call volume, and deployment that does not depend on GPU availability. That is the case for evaluating Lokutor as an ElevenLabs alternative.

What ElevenLabs costs, from its pricing page

Read on 25 September 2026:

  • Free: $0, 10k credits per month
  • Starter: $6 per month, 30k credits
  • Scale: $299 per month, 1.8M credits
  • Business: $990 per month, 6M credits, with low-latency TTS as low as 5 cents per minute at that tier

ElevenLabs Agents lists 8 cents per minute beyond a plan's allocation (read 23 September 2026). Credits, bundles, and what one minute buys differ between providers, so treat these as list rates rather than a like-for-like benchmark.

What Lokutor does differently

One price of about 2 cents per minute with speech recognition, the language model, and the voice included (1.4 cents on the Business plan). The free plan includes 60 minutes a month. The voice model, Versa, is a few tens of millions of parameters and runs on commodity CPUs, which is where the price comes from.

Latency from production traffic, single CPU thread: 139 ms median streaming time to first audio, 211 ms p90. At the full agent level, caller-observed time to first audio is 385 to 442 ms in the best case and about 660 ms typical.

Deployment runs on x86 and Arm CPUs: hosted API, private VPC, on-premises, or edge. In a measured AMD Genoa capacity test, one node handled about 19 concurrent calls, roughly $22 of infrastructure per concurrent stream per month.

Languages: Spanish, Catalan, Galician, Basque, English, Portuguese, French, Italian, and German. Zero-shot voice cloning from an eight-second reference.

Where ElevenLabs still wins

  • Voice variety and marketplace breadth
  • Dubbing and creative tooling
  • Ecosystem and brand recognition

If your product is content production, stay put. If your product answers phone calls and the GPU bill is becoming a problem, test us.

Evaluate in an hour

Get a free API key at app.lokutor.com, 60 minutes included. Docs at docs.lokutor.com. Run it against your current provider on the same script, cold and warm, and keep whichever wins your traffic.

Top comments (0)