Daniel Walsh gives both models matching briefs and high reasoning settings, then combines a model-based requirements review with hands-on inspection. The tasks include flood and driving simulations, a design editor, a bike-shop website, a Blender kitchen and a short launch film.
Daniel Walsh often favors Opus 5.5 for clearer interfaces and more coherent results, while noting functional and rendering faults in both models. A score can miss a serious problem: the driving demo retains flickering even when its automated assessment ranks it ahead. The flood examples are coding prototypes, not validated predictions of real flooding.
Daniel Walsh also compares runtime and token use, but flags interruptions, a usage limit and possible caching differences that complicate interpretation. The Blender kitchens show strong visual progress while retaining small construction defects. The takeaway is to test the actual requirements and user experience rather than treating one judge score or polished render as proof of readiness.
Watch on YouTube
