Qwen 3.8 Omni Flash

This article independently confirms Alibaba's launch of Qwen3.8-Omni-Flash as a native multimodal model that accepts text, images, audio, and video, removing the need for developers to set up separate systems per modality. The release ships with built-in thinking support, custom function calling, and web search, so the The piece highlights the 1M-token context window as the central differentiator, with Alibaba Cloud documentation showing a top input size of 991,808 tokens in non-thinking mode and 983,616 tokens in thinking mode. The authors add a useful caveat that a 1M-token window does not directly translate to unlimited audio or v

Alibaba's Qwen team has released Qwen3.8-Omni-Flash, described as its first omni-modal model built around agentic capabilities. According to the article, the model accepts text, images, audio, and video inputs and returns text only, combining audio-video understanding, reasoning, and tool use in a single model whose st Technically, Qwen3.8-Omni-Flash is built on the Qwen3.8-Flash-Next architecture (whose open-weights base shipped in August 2026) and offers a 1M-token context window, with QwenCloud listing 991K max input and 131K max output and a max reasoning length of 262K tokens. Thinking mode is on by default with reasoning effort

Qwen3.8-Omni-Flash is Alibaba's next-generation omnimodal model succeeding Qwen3.5-Omni-Plus, built around audio and video agents rather than just multimodal understanding. It processes text, image, audio, and video in one model with a 1M-token context window, while keeping text performance comparable to a text-only mo Headline benchmark gains include a +36.5 point improvement on WildClawBench-MM and a 98%+ reduction in the per-hour price of audio input versus the predecessor. Demonstrated agent capabilities include watching and listening, planning, and calling tools to deliver finished work such as music videos and film commentary.

Alibaba's Qwen team launched Qwen3.8-Omni-Flash on September 18, 2026, positioning it as a shift from understanding omnimodal content to agentic action — planning tasks, calling tools, and completing creative work. The model is a native omnimodal system accepting text, image, audio, and video inputs with a 1-million-to The launch ships with two new open-source components — Qwen-MM-Plugins (agent harness plugins) and Qwen-Live Harness (a real-time interaction runtime) — enabling workflows like Music2MV (generating music videos from song rhythm, mood, and timing), speaker-aware dialogue translation with voice cloning and dubbing, and f