01
What Realtime Plus provides
qwen-audio-3.1-realtime-plus is Alibaba's 3.1 end-to-end conversational audio model. It accepts text and audio, streams text and audio back, and is designed for live spoken applications rather than offline asset generation.
Alibaba currently documents a 262,144-token context window, maximum input length of 245,760 tokens and maximum output length of 16,384 tokens. Audio-history controls impose separate practical limits on retained spoken context.
02
WebSocket lifecycle and turn detection
The main integration path is event-driven WebSocket: configure the session, append microphone audio, consume response audio deltas, handle interruptions and close cleanly. Alibaba also documents AOQ and WebRTC integration paths.
Turn handling includes acoustic VAD, semantic smart-turn detection and push-to-talk. The current guide documents 16 kHz mono PCM input and 24 kHz mono PCM output and recommends small audio chunks for low-latency playback.
03
Function calling, web search and voice cloning
Realtime Plus supports function calling and built-in web search, but Alibaba currently states that the two cannot be enabled together. Design the session around the capability that matters rather than promising simultaneous native search and tools.
Cloned voices are supported through Alibaba's separate Voice Cloning API and a compatible target model. Use voice cloning only with appropriate rights and consent; a voice ID does not establish permission to imitate another person.
04
Regional pricing and service limits
For the Singapore international service, Alibaba currently lists $0.80 per million text-input tokens, $6.40 per million audio-input tokens, $6.40 per million text-output tokens and $24 per million audio-output tokens.
The Singapore documentation also lists 60 requests per minute and 100,000 tokens per minute. Beijing uses a separate price table, so there is no safe single global Realtime Plus price.
05
Conversation history and choosing the right endpoint
The realtime guide separately caps retained audio history by turns and cumulative audio duration. A large token context therefore should not be interpreted as unlimited raw audio history.
Use Realtime Plus when a user needs immediate spoken interaction. Use TTS-Next for produced audio assets and the ASR family when the application primarily needs transcription rather than an end-to-end conversational model.
Sources
Primary and supporting sources
Use the linked documentation to check current pricing, access and product behaviour. The dates above describe this article; they do not guarantee that every vendor claim remains current.