AICodeKing revisits Qwen 3.8 Max Preview after an update and reports improvements in frontend generation, agentic tool calling and long coding sessions. The earlier eight-task benchmark is presented as the channel's own test, not an independent industry ranking or a new benchmark run.
The presenter says the model now follows project structure more consistently and is less likely to omit required files inside a coding harness. The proposed explanation, a different checkpoint or inference deployment, is explicitly speculation because the underlying change has not been confirmed.
The video distinguishes access through Qwen Studio from heavier-use plans and notes that a free command-line client does not necessarily include free hosted inference. Planned open weights are described as a future release, not currently available weights. These access claims reflect the video's publication date.
Watch on YouTube
