AICodeKing evaluates Qwen 3.8 Max using an eight-task KingBench set that includes interface construction, geometry, SVG drawing, an archery game and mathematics. The reported 65 out of 80 score belongs to this creator's test set, not a broad independent assessment.
AICodeKing highlights strong individual outputs but also shows a weaker three-dimensional watch result. A longer Claude Code-style harness task fails to complete, with the demonstration leaving uncertainty about the contribution of the model, adapter and harness.
AICodeKing discusses the preview and anticipated open-weight release separately. Announced future availability is not the same as released weights, and this limited demonstration cannot establish universal superiority over proprietary coding models.
Watch on YouTube
