Qwen3.8 Omni Flash

Alibaba's Qwen team has released Qwen3.8-Omni-Flash, its first omni-modal model built around agentic capabilities. It accepts text, images, audio, and video as inputs and returns text, combining audio-video understanding, reasoning, and tool use in a single model. The model is positioned for workflows that understand content, plan tasks, execute tools, and deliver results. The model is built on the Qwen3.8-Flash-Next architecture, which shipped with open weights in August 2026. It offers a 1M-token context window, with QwenCloud listing 991K max input and 131K max output, plus a 262K max reasoning length. Thinking is enabled by default with reasoning effort set to xhigh. It is available as a hosted API on QwenCloud, Alibaba Cloud Model Studio, and Qwen Studio, with no open weights announced at launch.

Qwen3.8-Omni-Flash is Alibaba's native omnimodal model in the Flash tier, accepting text, image, audio, and video with a 1M-token context window. It targets audio and video agent workloads, with Alibaba claiming audio-visual performance close to Gemini 3.8 Flash and overall audio performance exceeding it. The model reportedly shows a 25%+ average improvement over Qwen3.5-Omni-Plus across 29 evaluations. DataCamp notes headline agent gains of 36.5 points on WildClawBench-MM and a reported 98%+ drop in per-hour audio input pricing. Text pricing is approximately $0.15 input and $0.47 output per 1M tokens, undercutting most Flash-tier rivals. DataCamp recommends it for audio- or video-heavy workloads, while noting Qwen3.8-Flash-Next scores marginally higher on pure text or code tasks.

BenchLM.ai presents Qwen3.8-Omni-Flash evaluation results across audio-visual agent benchmarks, audio-visual understanding, and audio tasks. Qwen3.8-Omni-Flash scored 71.0 on WildClawBench-MM (leading among four reported systems), 69.6 on UniClawBench, 36.8 on AgenticVBench, and 74.0 on OmniGAIA, generally trailing Gemini 3.8 Flash but outperforming Qwen3.5-Omni-Plus and Seed 2.0 Lite. On audio-visual reasoning benchmarks, Qwen3.8-Omni-Flash scored 63.4 on OmniVideoBench and 65.0 on Video-MME-v2, placing third among five reported systems. Using the Qwen Code agent harness raised its OmniVideoBench score to 67.8 while reducing tokens per query from 145,736 to 79,117. Metrics span accuracy for reasoning tasks and DER/cpWER/WER for ASR evaluations.