Overview
Qwen3.8-Omni-Flash is the omni-modal model Alibaba released on 18 September 2026. It accepts text, images, audio and video within a single 1M-token context window and returns text. Alibaba describes it as natively omni-modal, meaning one model handles all four input types rather than a pipeline of separate converters feeding a text model.
Alibaba Cloud's Model Studio documentation lists the model for audio and video understanding, meeting summaries and content analysis, with thinking mode, tool calling and web search supported. That combination is what makes it an agent model rather than a transcription endpoint: a single call can read a clip and then act on what it found.
Cost is the headline change over the previous generation. Alibaba reports that the API price per hour of audio input falls by more than 98% against Qwen3.5-Omni-Plus, and the price per hour of audio-visual input by more than 93%. Part of that comes from efficiency rather than discounting — Alibaba reports 51.8% fewer tokens on OmniVideoBench at comparable accuracy.
On quality, Alibaba reports an average improvement of more than 25% across 29 evaluations against Qwen3.5-Omni-Plus, including a 19.5-point average gain on agent benchmarks, and says the model approaches Gemini 3.8 Flash on audio-video benchmarks. Text performance is described as comparable to a text-only model of the same size.
The model is served from the cloud only — there is no weight download. Alongside it Alibaba published Qwen-MM-Plugins under Apache-2.0, a Python package that makes text-only agent harnesses accept audio and video input.
| Released | 2026-09-18 |
|---|---|
| License | Proprietary (hosted API) |
| Weights | API only |
| Context | 1M |
| Modalities | Text, Image, Audio, Video |
| Status | Generally available |
Strengths
- Text, image, audio and video handled by one model inside a 1M-token context window
- Audio input priced more than 98% below Qwen3.5-Omni-Plus per hour; audio-visual input more than 93% below
- 51.8% fewer tokens on OmniVideoBench at comparable accuracy, per Alibaba
- Thinking mode, tool calling and web search supported, so audio and video work can drive an agent loop
- Served from six Alibaba Cloud regions on an OpenAI-compatible endpoint
Best for
- Reach for it for meeting summaries and long-form audio-visual analysis where per-hour media cost dominates the bill
- Reach for it for video editing, commentary and audio-visual summarisation workflows that need tool calls, not just a transcript
- Reach for it when a job mixes long documents with recorded media and must stay inside one context window
- Reach for it when an existing OpenAI-compatible agent needs media input without rewriting the harness, using Qwen-MM-Plugins
How to access
| Provider | Model ID |
|---|---|
| Alibaba Cloud Model Studio ↗ | qwen3.8-omni-flash |
Qwen-Omni — every version
The full lineage of the Qwen-Omni line, newest first. Every version has its own page — click any to compare specs, benchmarks and pricing.
| Version | Released | Context | License |
|---|---|---|---|
| Qwen3.8-Omni-Flashcurrent | 2026-09-18 | 1M | Proprietary (hosted API) |
| Qwen3-Omni | 2025-09-22 | — | Apache-2.0 |
| Qwen2.5-Omni | 2025-03-26 | — | Open weights |
FAQ
What is Qwen3.8-Omni-Flash?
Qwen3.8-Omni-Flash is Alibaba's omni-modal model released on 18 September 2026. It takes text, images, audio and video inside a 1M-token context window and answers in text, with thinking mode, tool calling and web search supported. Alibaba serves it through Qwen Chat, QwenCloud and the Alibaba Cloud Model Studio API.
Are the weights open?
No. Qwen3.8-Omni-Flash is a hosted, API-only model and Alibaba has not published downloadable weights for it. The open artefact released alongside it is Qwen-MM-Plugins, an Apache-2.0 Python package on GitHub that adds audio and video handling to text-only agent harnesses.
Does it generate speech?
No. Alibaba Cloud's documentation lists Qwen3.8-Omni-Flash as taking audio and video in and returning text out. Speech output stays with the earlier Qwen3.5-Omni, which streams both speech and text but is offered in fewer regions. A project needing spoken replies has to pair Qwen3.8-Omni-Flash with a separate text-to-speech step.
How much cheaper is it than Qwen3.5-Omni-Plus?
Alibaba reports that the API price per hour of audio input falls by more than 98% and the price per hour of audio-visual input by more than 93% compared with Qwen3.5-Omni-Plus. Alibaba attributes part of that to efficiency rather than discounting, reporting 51.8% fewer tokens consumed on OmniVideoBench at comparable accuracy.
Which regions is it available in?
Alibaba Cloud lists six regions for Qwen3.8-Omni-Flash: Beijing, Singapore, Hong Kong, Tokyo, Frankfurt and Virginia. Each region requires its own API key and workspace URL, so an application must call the endpoint for the region where it is registered. The endpoint is OpenAI-compatible and supports streaming.