Alibaba released the multimodal model Qwen 3.8 Omni Flash

Company Alibaba introduced a new native multimodal model Qwen 3.8 Omni Flash, focused on high-performance data processing, and the model is already available for use on the Qianwen AI platform.

A key feature of the architecture is the extended context window of 1 million tokens, and according to the developers, the performance of the new model exceeds the previous version Qwen 3.5 Omni Plus by 25% in standard benchmarks.

Special attention in the update is paid to optimizing the costs of processing audio data, with the company's engineers significantly reducing the cost of incoming audio streams, making the model more cost-effective for integration into real-time speech recognition and voice interface systems.

Technical details of the implementation and the first community reviews of the model's operation are discussed on the Hacker News platform, with developers noting that Qwen 3.8 Omni Flash demonstrates increased accuracy in analyzing long documents and multimedia files.

The release of this model continues Alibaba's strategy of developing the Qwen line, aimed at lowering the entry threshold for corporate clients, and increasing the context window to 1 million tokens allows the system to be used for analyzing large data archives without the need for pre-segmentation.