Qwen 3.8‑27B Open‑Source Release: Boosted Inference, Coding, and Agent Capabilities

Qwen 3.8‑27B, a 27 billion‑parameter large language model derived from the Qwen 3.5 architecture, abandons the traditional "larger‑is‑better" scaling route and instead focuses on comprehensive upgrades to inference, code generation, and autonomous‑agent capabilities.

In official evaluations the model reaches SWE‑bench Verified 72.4 and LiveCodeBench v6 80.7**, demonstrating programming proficiency close to larger models. As a native vision‑language model it scores OmniDocBench 1.5 88.9 and VideoMME 87.0**, showing strong image, document, and long‑video understanding. Agent‑oriented tests report ScreenSpot Pro 70.3 and OSWorld‑Verified 56.2**, indicating improved task planning, tool use, and environment feedback.

The model also provides a flexible reasoning‑depth control mechanism, allowing users to balance performance and efficiency according to their needs.

HyperAI (hyper.ai) now hosts an online tutorial titled "Qwen3.8‑27B‑FP8: Native Vision‑Language Model" that guides developers through end‑to‑end usage.

Visit the HyperAI homepage, open the Tutorial section, select "Qwen3.8‑27B‑FP8: Native Vision‑Language Model", and click "Run this tutorial".

On the tutorial page, click the top‑right Clone button to copy the notebook repository into your own container.

Choose the NVIDIA RTX 5090‑2 instance and the vllm image, then press "Continue job execution".

Wait until the job status changes to Running , then open the Jupyter workspace via "Open Workspace".

Inside the workspace, open the README, click Run , and after execution launch Open WebUI with the provided command; finally, click the displayed API address to start interacting with the local model.

These steps enable rapid experimentation with Qwen 3.8‑27B's multimodal and agent features without requiring local model setup.