Alibaba has launched Qwen3.8-Omni-Flash, an omni-modal model designed for audio-video understanding and tool usage. The model is built on the Qwen3.8-Flash-Next architecture and is currently available via API on QwenCloud, Alibaba Cloud Model Studio, and Qwen Studio. Open weights for this version have not been released.
The model features a 1 million token context window, with a maximum input of 991,000 tokens and a maximum output of 131,000 tokens. It supports function calling, web search, and context caching. While the model processes audio and video, its output is strictly text-based.
According to internal research, the model uses an agentic approach to process video, selecting specific segments to analyze rather than scanning entire files. This method reportedly increased accuracy on OmniVideoBench from 63.4 to 67.8 while reducing token consumption by 45.7%.
Pricing for the API is set at $0.15 per 1 million input tokens and $0.47 per 1 million output tokens. Alibaba reports significant cost reductions compared to the Qwen3.5-Omni-Plus model, citing a 98% decrease in audio input costs and a 93% reduction for audio-visual input.
To support multimodal integration, the company open-sourced Qwen-MM-Plugins under the Apache-2.0 license. These plugins allow developers to connect the model with various coding environments, including Claude Code and Gemini CLI, to facilitate local image and video frame processing.
