# Alibaba releases Qwen3.8-Omni-Flash, a 1M-token omni-modal model with agentic audio‑video

Digest AI · Generative AI & Models · published 2026-09-18T08:40:37Z · updated 2026-09-20T06:46:08Z

Canonical: https://digestai.news/story/alibaba-releases-qwen3-8-omni-flash-a-1m-token-omni-modal-model-with-a

## Summary

Alibaba’s Qwen team announced the launch of Qwen3.8-Omni-Flash, its first omni‑modal model that handles text, images, audio and video and returns text. The model runs as a hosted API on QwenCloud, Alibaba Cloud Model Studio and Qwen Studio; no open‑weight version was released, so self‑hosting is not possible. It is built on the Qwen3.8-Flash-Next architecture and offers a 1 M‑token context window, with a maximum of 991 K input tokens, 131 K output tokens and 262 K tokens for reasoning. Pricing listed by QwenCloud is $0.15 per 1 M input tokens, $0.47 per 1 M output tokens, and $0.016 per 1 M cached tokens.

The research team reports that the agentic perception approach improves OmniVideoBench accuracy from 63.4 to 67.8 while reducing token usage by about 45.7 %. Across 29 evaluations the model averages more than a 25 % score gain over Qwen3.5-Omni-Plus, with notable improvements on WildClawBench‑MM (+36.5 points) and UniClawBench (69.6). All benchmark figures are from Qwen and have not been independently verified. The team also open‑sourced Qwen‑MM‑Plugins and Qwen‑Live Harness under Apache‑2.0 to help developers build multimodal agents, though audio must still be routed through the API rather than fed directly.

## Key points

- Qwen3.8-Omni-Flash supports text, image, audio, video input and returns text via a hosted API.
- Model offers 1 M‑token context (991 K input, 131 K output) and costs $0.15 per 1 M input tokens, $0.47 per 1 M output tokens.
- Agentic video perception raises OmniVideoBench accuracy from 63.4 to 67.8 while cutting token use by 45.7%.

## Why it matters

A 1 M‑token omni‑modal model with built‑in tool use lets developers build richer AI agents for long video and audio analysis at lower token cost, expanding practical multimodal applications.

## Sources

1. [Alibaba Qwen Releases Qwen3.8-Omni-Flash: A 1M-Context Omni-Modal Model Built Around Agentic Audio-Video Understanding and Tool Use](https://marktechpost.com/2026/09/18/alibaba-qwen-releases-qwen3-8-omni-flash) (MarkTechPost, 2026-09-18)
2. [Alibaba Qwen Team Releases Qwen3.8-LiveTranslate: A Real-Time Interpretation Model That Cuts Average Lag to 2.3 Seconds Across 60 Languages](https://marktechpost.com/2026/09/19/alibaba-qwen-team-releases-qwen3-8-livetranslate) (MarkTechPost, 2026-09-20)
3. [Qwen3.8-Omni-Flash undercuts Google's Gemini Flash pricing while matching its multimodal benchmarks](https://the-decoder.com/qwen3-8-omni-flash-undercuts-gemini-flash-pricing-while-matching-its-multimodal-benchmarks) (The Decoder, 2026-09-19)

Part of the developing story: [Alibaba Expands Qwen AI Ecosystem With New Hardware](https://digestai.news/thread/alibaba-releases-qwen3-8-omni-flash-a-1m-token-omni-modal-model-with-agentic) (4 stories)

## Cite

Digest AI, "Alibaba releases Qwen3.8-Omni-Flash, a 1M-token omni-modal model with agentic audio‑video", 18 September 2026, https://digestai.news/story/alibaba-releases-qwen3-8-omni-flash-a-1m-token-omni-modal-model-with-a

---

Written by Digest AI's editorial model from the linked sources; the sources are the record. Headlines, digests and key points are written by Digest AI and may be quoted with a link to the story page. Linked articles belong to their publishers. Terms: https://digestai.news/terms#reuse
JSON: https://digestai.news/story/alibaba-releases-qwen3-8-omni-flash-a-1m-token-omni-modal-model-with-a.json