September 19, 2026
One AI Handles Text, Photos, Audio, and Video: Qwen3.8-Omni-Flash Supports 1 Million Tokens
Alibaba released Qwen3.8-Omni-Flash with a 1 million-token context window on September 19, 2026. The model accepts text, images, audio, and video, and returns text through Chat Completions and Responses. In international deployment, input costs $0.15, output $0.47, and caching $0.016 per 1 million tokens.
Alibaba previously compared multimodal scenarios with Qwen3.5-Omni-Plus. The company now says Qwen3.8-Omni-Flash outperforms it by more than 26% on average across 30 evaluations. On OmniVideoBench, the score rose from 63.4 to 67.8, while token consumption fell from 145736 to 79117, or roughly 45.7%.
Limits disclosed. A request without thinking can contain up to 991808 tokens, and with thinking up to 983616. A response can be up to 131072 tokens long. Through public URLs, the model accepts up to 2048 files of up to 2 GB each, audio up to 3 hours long, and video up to 2 hours long. The model is available in six regions, including Singapore, Germany, and US.
Work scenario. In a meeting lasting up to an hour, the model separates speakers, transcribes speech, links remarks to participants, and produces minutes with a task list. Through tools, it can send emails, create tasks, and write code.
Integration. In Model Studio, create an API key, install `openai`, set `DASHSCOPE_API_KEY` and `DASHSCOPE_BASE_URL`, then call `qwen3.8-omni-flash` through Chat Completions. Alibaba also expanded Qwen-MM-Plugins for long multimodal workflows and opened Qwen-Live Harness for continuous real-time interaction.
Alibaba expanded Qwen-MM-Plugins and opened Qwen-Live Harness for continuous real-time multimodal interaction.
Source
