Qwen launches Qwen3.8-Omni-Flash for audio-video agents
Alibaba's Qwen team has released Qwen3.8-Omni-Flash, an omnimodal model with a 1M-token context that is built for agentic audio and video work and cuts audio-input API prices by more than 98%.
The Qwen team announced Qwen3.8-Omni-Flash, a next-generation native omnimodal model. Qwen says the model is meant to move omnimodal systems beyond understanding content and toward planning tasks, calling tools and finishing work such as video editing, music video creation, film commentary, audio-visual summarization and real-time conversation.
The model supports a 1M-token context window. Qwen says its text performance stays comparable to a text-only model of the same size. Across 29 audio, audio-visual and agent benchmarks, its average score is more than 25% higher than Qwen3.5-Omni-Plus. Qwen estimates that the API price per hour of audio input is more than 98% lower, and per hour of audio-visual input more than 93% lower. These hourly figures are extrapolated from the cost of two-minute clips, with video sampled at 720p and 1 fps.
Reported gains include 36.5 points on WildClawBench-MM, 22.3 points on AgenticVBench, 8.3 on LongAudioSpan and 9.6 on OmniVideoBench, plus a score of 69.6 on UniClawBench. On AliMeeting multi-speaker recognition, DER falls from 88.11 to 3.35 and cpWER from 89.61 to 17.18. Qwen says audio-visual performance is close to Gemini 3.8 Flash and overall audio performance is above it. These are Qwen's own results and have not been independently verified.
Alongside the model, Qwen expanded Qwen-MM-Plugins with on-demand perception, tool use and workflow execution for long-form media. It also open-sourced Qwen-Live Harness, a runtime for continuous, real-time omnimodal interaction.
The source excerpt does not say whether the model weights will be released, give exact per-token prices, or say when the model will be available.