Alibaba's Qwen team has released Qwen3.8-LiveTranslate, a hosted model that translates speech while a person is still talking rather than waiting for them to finish. The model listens to live audio, can optionally use video frames for extra context, and streams back translated text and speech during the conversation.
The headline improvement is latency. Qwen3.8-LiveTranslate lowers average lag, measured by length-adaptive average lagging (LAAL), from 2.8 seconds in the previous Qwen3.5-LiveTranslate to 2.3 seconds. New capabilities include real-time speaker separation in multi-person conversations, synchronised display of source and translated text, and the use of earlier context in a conversation to disambiguate names and technical terms it has already heard.
Language coverage stays the same as the previous generation: the model understands around 60 languages and speaks 29 of them. It is API-only, served over a WebSocket Realtime API under the model ID qwen3.8-livetranslate-flash-realtime, with no weights released for self-hosting.
Reported pricing varies by source, with figures around $7.50 per million audio input tokens and roughly $20 per million output tokens cited for the API; Alibaba Cloud's own pricing page should be checked before budgeting around these numbers.
The release targets live interpretation use cases such as conference calls, customer support and events, where a translation delay of several seconds can otherwise break the flow of conversation.
Source: MarkTechPost
