Alibaba has launched Qwen3.8-Omni-Flash, a native multimodal model capable of processing text, images, audio, and video inputs with support for a 1M token context window. The model is now available on the Qwen AI platform.
Performance metrics show a 26% average improvement across 30 benchmarks compared to the previous Qwen3.5-Omni-Plus version. Specific gains include a 36.5-point increase in WildClawBench-MM and a 22.3-point rise in AgenticVBench.
The company reports that audio capabilities now exceed those of Gemini 3.8 Flash. Pricing for API usage has been significantly reduced, with audio input costs dropping by over 98% and audio-video input costs by over 93% per hour.
New tools include the open-source Qwen-Live Harness for real-time interaction and expanded Qwen-MM-Plugins. The model supports complex tasks like long-form video analysis, meeting transcription, and automated content production.
A real-time version, Qwen3.8-Omni-Flash-Realtime, features spatial audio-visual integration to identify sound source direction and distance. It also supports real-time language practice and handles non-standard speech patterns.
