Pat Simmons begins with Google's demonstrations and benchmark claims, then evaluates four prompts using anonymized output columns before revealing the model identities. The tasks cover a browser-based product model, a short launch video, a car-soccer game and an animated website recreation. The process tests specific creative workloads rather than establishing a universal coding ranking.
Pat Simmons pays particular attention to material realism, liquid motion, camera transitions, game physics and scrolling interactions. In this set of outputs, the presenter favors Opus 5.5, while Gemini 4 Argon often finishes more cheaply but misses visual or behavioral requirements. Strong individual details do not necessarily compensate for broken transitions or weak overall coherence.
Pat Simmons reports build durations and API-equivalent costs alongside the results, noting that some runs consume much more time and many more tokens. The closing assessment acknowledges the limited, unusually demanding test set and leaves everyday knowledge-work performance unresolved. Release-period pricing, promotional discounts and subjective rankings are not presented as current purchasing advice.
Watch on YouTube
