Kokoro 82M vs Orpheus 3B
Two text-to-speech models compared on price, context, capabilities and measured speed. External prices are public list rates verified 2026-09-16; Inference APIs prices are live.
| Kokoro 82M | Orpheus 3B | |
|---|---|---|
| Provider | Hexgrad | Canopy Labs |
| Model id | hexgrad/Kokoro-82M | canopylabs/orpheus-3b-0.1-ft |
| Price | $5.20 / 1M characters | $19.50 / 1M characters |
| Capabilities | Text to speech, 54 voices, MP3 / WAV, 8 languages | Text to speech, Expressive, 8 voices, English |
| Weights | Open | Open |
| Availability | Available on Inference APIs | Available on Inference APIs |
| Rate limits | No daily caps; pay per request | No daily caps; pay per request |
| 300 characters → audio | 0.84 s | 58.5 s |
Cost for 1M characters (≈ 11 hours of speech)
Kokoro 82M is about 73% cheaper for this workload at list price.
Measured speed
Speed figures for Inference APIs models are medians of three runs from a European client on 2026-09-16, against the public endpoint, using a ~120-word generation prompt (chat), a 300-character paragraph (speech) or a -second clip (transcription). External models are not measured here; treat "not measured" as unknown, not slow.
When to pick which
- Kokoro 82M — Fast, natural open-weight text-to-speech with 54 voices across 8 languages, on the OpenAI /v1/audio/speech endpoint.
- Orpheus 3B — Expressive Llama-based text-to-speech with emotive tags and eight English voices.
Try Kokoro 82M
Same OpenAI request shape; the change is the base URL and key. Full parameters, aliases and pricing on the model page.
External prices and limits come from the linked provider pages and were verified on 2026-09-16. Spotted a change? Tell us.