Alibaba Cloud released the Qwen3.8 large language model family in August 2026. The series includes the Qwen3.8-Max model with 2.4 trillion parameters, as well as the Qwen3.8-27B version designed for local execution on personal computers.
The Qwen3.8-27B model features 27 billion parameters and a context window of 262,144 tokens. According to the Intelligence Index benchmark by Artificial Analysis, the model scored 41 points, outperforming its predecessor, Qwen3.6-27B, which scored 29 points.
The model is distributed under the Apache 2.0 license. In a 4-bit quantization format, the file size is less than 20 GB, allowing it to fit into the VRAM of consumer-grade graphics cards like the RTX 3090 or 4090, or the unified memory of modern ARM-based Apple Mac computers.
Users can run the model locally using the Unsloth Desktop environment, which utilizes the llama.cpp inference engine. The software automatically suggests optimal quantization settings based on the available system VRAM and RAM. As of mid-September 2026, the Qwen3.8-27B model had been downloaded 10.5 million times via Unsloth Desktop.
