Qwen Launches Qwen3.8-Omni-Flash, a 1M-Context Omnimodal Model Built for Agents
Alibaba’s Qwen team released Qwen3.8-Omni-Flash on September 18, 2026, a native omnimodal model designed to move beyond passively describing audio and video toward actually acting on it — planning, calling tools, and completing creative work from what it sees and hears.
Details
- Broad input, huge context: accepts text, image, audio, and video with a 1M-token context window, enough for up to an hour of continuous audio-visual content in one session
- Big jump over predecessor: scores 25% higher on average across 29 evaluations versus Qwen3.5-Omni-Plus, with audio-visual performance close to Gemini 3.8 Flash and audio performance that exceeds it
- Sharp price cuts: audio-input API pricing drops 98%, and audio-visual input pricing drops 93%, compared to the previous generation
- Wide language coverage: recognizes speech in 74 languages plus 39 Chinese dialects, and generates speech in 29 languages plus 7 dialects
- Agentic production features: long-form video captioning, meeting transcription with speaker recognition, music-video creation synced to lyrics and rhythm, automated video translation with voice preservation, and tools like Video2Note and Omni Skill Creator
- Real-time option: a Qwen3.8-Omni-Flash-Realtime variant supports continuous audio-visual conversation over WebSocket/WebRTC, including spatial audio awareness and accent-aware speaking practice
- Availability: live now on the Qianwen AI Platform via an OpenAI-compatible Chat Completions API, alongside open-source tooling (Qwen-MM-Plugins, Qwen-Live Harness, Qwen Code)
What happened next
Qwen3.8-Omni-Flash lands as Alibaba pushes its omnimodal line toward agentic use rather than pure understanding, following close on the heels of its Qwen3.8-Max text model family. The release sets up two fast follow-ups from the same team within days — Qwen3.8-LiveTranslate for real-time interpretation and the open-weight Qwen-Image-2.1 — pointing to an aggressive September release cadence across Qwen’s multimodal lineup.