Qwen3.8 Omni Flash

Alibaba Cloud / Qwen Team · proprietary · qwen3.8-omni-flash

Qwen3.8-Omni-Flash is Alibaba/Qwen's native omnimodal API model (id qwen3.8-omni-flash) for agentic audio-visual task execution. Inputs: text, image, audio, and video; output: text. Supports a 1M-token context window, reasoning_effort (xhigh/medium/low), and OpenAI-compatible Chat Completions / Responses APIs on QwenCloud (DashScope). Released 2026-09-18 as API-only (no open weights on this card). Vendor paper/blog highlights include WildClawBench-MM 71.0, UniClawBench 69.6, LongAudioSpan Accuracy 82.7, OmniVideoBench 63.4 static / 67.8 agentic (Qwen Code), MMAU 81.8, MMAR 79.8, MMSU 82.1, VoiceBench 91.6, OmniGAIA 74.0, DailyOmni 85.1. Catalog rows are limited to existing benchmark ids; UniClawBench, LongAudioSpan, OmniVideoBench, MMAR, and MMSU have no catalog ids yet and are noted here only. No first-party provider price row in this repo — available_in_zeroeval is false.