Qwen3.8-Omni-Flash Launches With 1M Context and Advanced Audio-Visual Agents

Qwen has unveiled Qwen3.8-Omni-Flash, its latest native omnimodal model capable of processing text, images, audio, and video. This advanced model features a 1-million-token context window and can plan tasks, utilize tools, and execute multi-step operations. Designed to enhance workflows such as video editing, music video creation, film commentary, audio-visual summarization, and real-time conversations, Qwen3.8-Omni-Flash also boasts significant improvements over its predecessor.

With an average score surpassing Qwen3.5-Omni-Plus by more than 25% across 29 evaluations, Qwen claims a 98% reduction in audio input costs and a 93% decrease in audio-visual input costs under its pricing structure. The model's audio-visual performance closely mirrors Gemini 3.8 Flash while outperforming it in overall audio capabilities, according to Qwen's own testing.

Notably, Qwen3.8-Omni-Flash excels at analyzing specific elements within recordings, such as characters, camera shots, lighting, and sound, based on user requests. This capability, combined with its ability to process up to one hour of audio-visual meeting input, enables the model to identify speakers, transcribe discussions, generate minutes, extract action items, and analyze project risks.

When integrated with external tools, the model can perform tasks like sending emails, organizing tasks, or initiating coding based on meeting requirements.

Additionally, Qwen3.8-Omni-Flash-Realtime allows for real-time audio and video processing, enabling the model to determine the origin of sounds in a given environment and assist with spatial navigation. Speech recognition is supported in 74 languages, including Urdu and Punjabi, while speech generation covers 29 languages. Furthermore, Qwen has expanded Qwen-MM-Plugins and made Qwen-Live Harness available as open-source resources for multimodal agent workflows, long-term memory, task delegation, and real-time interaction.