Bijan Bowen tests Opus 5 on a browser-style operating system, a C++ skateboarding game, a reconstructed television set, a product site, a first-person subway game and an interactive city timeline. The tasks emphasize complex visual output rather than short benchmark answers.
The model's strongest results come from textures, spatial detail and sustained iterationAn agent loop repeatedly observes a task's state, chooses an action, checks the result and decides whether to continue or stop.. Several projects feel more completeCode quality describes how well software code meets relevant needs for correctness, clarity, maintainability and safe change in its intended context. than typical one-shot generationsCode generation uses AI or another automated system to create source code from instructions, examples, schemas, or higher-level specifications., and Opus 5 often keeps refining the implementation instead of stopping at a superficial mock-up.
The review also finds uneven execution. Some requested hero elements or pedestrians never appear, and the model can take a long time to finish ambitious builds. The practical conclusion is that Opus 5 raises the visual ceiling, but its output still needs direct inspection and testingEvaluation is the systematic assessment of a system's behaviour or results against explicit criteria..
Watch on YouTube




