Alibaba's Qwen3.8-Omni-Flash Open-Sources Plugins That Bolt Audio and Video Onto Claude Code, Codex and Gemini CLI
Qwen released Qwen3.8-Omni-Flash on 18 September 2026, a 1M-context native omnimodal model taking text, image, audio and video, claiming a 26%-plus average gain over Qwen3.5-Omni-Plus across 30 evaluations, +36.5 points on WildClawBench-MM, 69.6 on UniClawBench, and AliMeeting DER/cpWER falling from 88.11/89.61 to 3.35/17.18. Its Agentic Understanding mode raises OmniVideoBench accuracy from 63.4 to 67.8 while cutting token consumption from 145,736 to 79,117, a 45.7% reduction. Weights are not released and the model is API-only, but Qwen open-sourced Qwen-MM-Plugins, which drop image, audio, long-video and video-editing capability into Claude Code, Codex, Gemini CLI and Qwen Code, plus a Qwen-Live Harness real-time scheduling layer.
↳ Follow the thread