stupidmeter
An AI model benchmarking tool with a deliberately retro UI. Ranks models by their "stupid level" -- presumably testing for common failure modes, logical errors, and hallucination patterns. The retro design aesthetic is the hook.
Key Quotes
(Page is JavaScript-rendered and couldn't be fully fetched.)
Key Themes
The approach of measuring stupidity rather than intelligence is a useful inversion. Most benchmarks measure what models can do; measuring what they get wrong is arguably more useful for practical deployment decisions. The framing connects to the eval philosophy in Demystifying Evals for AI Agents -- testing for failure modes, not just capabilities.
Critical Analysis
Without being able to fetch the full content, it's hard to assess the methodology. The "retro UI is briefly cool again" annotation suggests this is as much a design statement as a serious benchmarking tool. The site likely joins a crowded field of model comparison tools, but the framing (stupid level rather than smart level) and the aesthetic distinguish it.
The real question with any model ranking tool: how quickly does it update when new models ship, and how reproducible are the benchmarks? Most model ranking sites become stale within weeks of launching.
Sources: stupidmeter Last updated: 2026-05-14