Developer uses Claude Opus 5.5 to generate 500 web clocks

By Splitfeed EditorialLast updated Oct 18, 2026, 06:00 UTC

On October 18, 2026, a developer posted to Hacker News a single self-contained HTML file, 6.3 megabytes in size, containing 500 distinct clock designs generated by Claude Opus 5.5. The work accumulated across multiple chat sessions in the Claude interface, with Claude itself producing a GitHub-hosted summary of the prompting process and the overall page design. The developer framed the project as a personal benchmark for evaluating frontier AI models, explicitly comparing it to Simon Willison's pelican-on-a-bicycle test, a creative-output prompt widely used as an informal gauge of model capability.

The project sits at the intersection of two ongoing debates: how useful creative or aesthetic tasks are as proxies for AI capability, and whether the latest generation of large models has meaningfully improved on its predecessors. Generating 500 visually distinct objects in a domain as constrained as clock faces is a reasonable stress test for variety and originality, not just correctness. The developer's stated satisfaction with Claude Opus 5.5's output over other tested models adds a comparative dimension, though the specific models it was measured against are not named in the available material.

None of the three camps have published reactions to this story yet. The Pro-AI camp would typically treat a demonstration like this as evidence of genuine creative breadth in current models, pointing to the volume and variety as signs that AI has moved well beyond template-level generation. The Anti-AI camp would likely question whether 500 algorithmically produced clock widgets represent meaningful creativity or simply statistical variation dressed up as design, and might argue the benchmark flatters the tool more than it tests it. The Middle Ground camp would probably welcome informal benchmarks of this kind as useful complements to formal evals, while cautioning that one developer's subjective quality rating is hard to generalize.

The natural next development to watch is whether other developers replicate the clock benchmark with competing models and publish comparative results, which would give the informal metric more weight or expose its limits.