The open-weight 82M-parameter TTS model that delivers studio-grade voices on a laptop CPU.
Kokoro is the open-weight text-to-speech model that punched far above its size class when it dropped in early 2025. At just 82 million parameters it tops community TTS quality arenas, ships under Apache-2.0, and runs in real time on a laptop CPU — no GPU, no API key, no per-character bill. Builders reach for it when they need narration, agents, or audiobooks without cloud pricing.
Who it's for: Indie developers, accessibility builders, and anyone producing hours of narration who wants cloud-quality voices without cloud fees.
Tiny enough to run on CPU in real time — deploys anywhere.
Rated alongside or above paid cloud TTS in blind listening tests.
American/British English plus Japanese, Chinese, French, and more.
Generate minutes of audio in seconds on commodity hardware.
The best free TTS you can self-host, full stop. For character work or cloned voices, ElevenLabs still leads — but for bulk narration on a budget, Kokoro is unbeatable.