Prices are per million input characters. Speed is the median over real ModelRelay requests, in characters synthesized per second. Voice quality is the Artificial Analysis TTS arena Elo, from blind listener votes. Models are listed rather than ranked until there are enough of them to compare.
| Model | Price | Speed chars/s, ModelRelay median | First audio streamed, ModelRelay median | Voice quality AA arena Elo |
|---|---|---|---|---|
| Gemini 3.8 Flash TTS Retained | $9.00/1M audio tokens + $0.50/1M text tokens in, $0.00225 per 10 s of audio | 29 chars/s | 675 ms | 1275 ±15 |
| Gemini 3.8 Flash-Lite TTS Retained | $6.00/1M audio tokens + $0.50/1M text tokens in, $0.0015 per 10 s of audio | — | — | 1248 ±14 |
Use ModelRelay's Responses API directly, or point an existing OpenAI or Anthropic SDK at the compatible endpoints.
curl https://api.modelrelay.ai/api/v1/responses \
-H "Authorization: Bearer $MODELRELAY_API_KEY" \
-H "Content-Type: application/json" \
-d '{"model":"claude-sonnet-5-5","input":"Hello"}'Change the model. Keep everything else.
Open to everyone. Sign in with GitHub or Google to get started.