Skip to main content
2026/9/14 13:30 (UTC+8) · New Model · OpenAI / Alibaba 🚀 Four Realtime voice models are out of private beta — select the default group on your key and call directly. Test, explore and integrate freely All four share one endpoint, wss://api.apiyi.com/v1/realtime?model=<model>, and one key; the default, VIP and SVIP groups all carry them:
  • gpt-realtime-2.1 / gpt-realtime-2.1-mini: OpenAI Realtime GA protocol passed through as is (including the newer reasoning.effort and semantic_vad fields); audio $32 / $64 and $10 / $20 per 1M tokens
  • qwen3.5-omni-plus-realtime / qwen3.5-omni-flash-realtime: Alibaba Model Studio protocol, tuned for Chinese; text+audio output $41.26 and $14.71 per 1M tokens
Text, audio and image tokens are billed per token at vendor list prices, group ratio 1. The 9/14 re-test ran the two GA models at 20 and 40 concurrent sessions, 120 of 120 succeeded, handshake and first text delta p50 about 1 second, and a session stayed alive through 5 minutes of silence. Two things to know up front: cache hits are reported faithfully in usage, but APIYI currently bills cached input at the full text-input rate (a separate notice will follow once the discount is live); and only direct WebSocket is supported — client_secrets, WebRTC and SIP endpoints return 404, so browser and mobile clients should go through a backend relay. Field comparison for both protocols, the text smoke test and the full list of known limitations are in the Realtime voice overview — if anything is missing or differs from what you measure, tell us.
← Back to Live Updates · 📚 Monthly archive