Alibaba Group unveiled its next-generation native omni-modal model, "Qwen3.8-Omni-Flash," designed to process text, images, audio, and video within a single model with a 1 million token context window. Compared to the previous-generation Qwen3.5-Omni-Plus, API pricing for one hour of audio input has been reduced by more than 98%, while the price for one hour of combined audio and video input has dropped by over 93%. The average score across 29 benchmarks improved by more than 25% versus the previous generation, with scores of 71.0 on WildClawBench-MM (a 36.5-point increase), 69.6 on UniClawBench, 82.7 on LongAudioSpan, 63.4 on OmniVideoBench, 28.2 on OmniCap-IF, and 89.7 on AliMeeting. Alibaba stated the model outperforms Google's "Gemini 3.8 Flash" on multiple benchmarks, achieving near-parity on audio-video performance while surpassing it on overall audio performance. Speech recognition supports 74 languages and speech generation covers 29 languages, including Japanese in both, plus Cantonese, Uyghur, and Māori in recognition. A key technical highlight is an "agentic understanding" mechanism that autonomously determines "where to look and what to listen to" for long-form audio and video.