What changed
Qwen’s repository lists Qwen3.8-2.4T-A95B on August 12 and Qwen3.8-27B on August 14. The 27B model card describes native vision-language capabilities and flexible thinking control.
The everyday takeaway
The specific download matters. Our local tests use a quantized 27B model, not the much larger model in the same family. Results from one should not be attributed to the other. The family name cannot tell you how much memory you need or how quickly a particular computer will answer your prompt.
What we have verified
The Qwen repository dates the 27B release to August 14. Our recorded local samples are in the Prompt Lab, with the recorded model identifier and available recording conditions.