Alibaba’s Qwen3.8-27B brings advanced AI tasks to local hardware

Alibaba has released Qwen3.8-27B, an open-weight AI model designed to handle coding, reasoning, images and video on local machines. Carl Franzen reports for VentureBeat that the 27-billion-parameter model is available under the Apache 2.0 licence, allowing companies and developers to download, adapt and host it themselves.

The release has attracted attention because of its balance between performance and hardware needs. At 16-bit precision, the model requires about 56GB of GPU memory. An FP8 version needs roughly 28GB. With 4-bit quantisation, however, the model file falls to around 17GB, making it usable on high-end consumer computers and some well-equipped laptops.

Alibaba reports strong results on coding and agent benchmarks, including SWE-bench Pro and LiveCodeBench. Independent testing has also produced notable results. Artificial Analysis assigns Qwen3.8-27B a score of 52 on its Intelligence Index, matching the score it gives OpenAI’s GPT-5.6 Luna at its highest reasoning setting. Such comparisons should be treated carefully, as test methods and configurations differ.

Capability comes with a speed cost

The model’s default reasoning mode can be slow and expensive in terms of computing resources. Artificial Analysis says Qwen3.8-27B generated substantially more output tokens than comparable open models in its tests. Developer Simon Willison reports that a simple image-generation request took 21 minutes at the model’s highest reasoning setting.

Willison recommends using low or no reasoning for routine work. He also found that Multi-Token Prediction, a feature supported by the model, improved inference speed when enabled through llama.cpp.

For organisations, the main question is not whether Qwen3.8-27B replaces every leading cloud model. It is whether local deployment is sufficient for tasks such as code assistance, document analysis and visual workflows. Running the model within an organisation’s infrastructure can give teams more control over data, costs and governance.