Alibaba (Qwen)

Qwen-Audio-3.1-Realtime

Released Sep 22, 2026

Alibaba's current full-duplex realtime voice model family on Model Studio, in Plus and Flash tiers. Alibaba points GPT-Live-1 and Gemini 3.8 Live users to the Plus tier.

Price per minute: —
Pricing: qwen-audio-3.1-realtime-plus per 1M tokens, Singapore: text in $0.80, audio in $6.40, text out $6.40, audio out $24 (audio-only billing when both are returned); Beijing: $0.688 / $5.501 / $5.501 / $20.628. Flash pricing not found.
Inputs: audio, text
Tool calling: Yes
Languages: 14
Voices: —
Big Bench Audio: —
Access: Generally available
Licence: Proprietary

Strengths & weaknesses

Strengths

14 languages plus 21 Chinese varieties
Function calling, plus web search on Flash
Voice cloning via the Voice Cloning API
262K context on Plus

Weaknesses

On Plus, function calling and web search can't be enabled together
Not yet on the AA leaderboard

Not notable: not on AA yet (the 3.0 generation is). releaseDate is the Apsara Conference press release naming Qwen-Audio-3.1-Realtime (2026-09-22). The Plus and Flash tiers are two model IDs kept in one entry; split them if preferred. Flash pricing isn't on the fetched pages. Alibaba's speech-to-speech page maps GPT-Live 1 and Gemini 3.8 Live to the 3.1 Plus tier, and GPT-Realtime-2.1 Mini to Flash. Voices: default longanqian_v3.1 plus the qwen-audio-3.1 TTS voice lists; no count is given.

Links

API model strings

alibabaqwen-audio-3.1-realtime-plus
alibabaqwen-audio-3.1-realtime-flash