My personal AI benchmark: "Generate an SVG of a frog with a Habsburg jaw."

AI frogs got royal chins and the commenters are absolutely losing it

TLDR: A goofy benchmark asked 14 AI models to draw a frog with a famous royal-style underbite, and all 42 attempts produced something. The real show was the comments: people argued over whether the models were smart for getting the jaw idea or ridiculous for turning frogs into bizarre little monarchs.

The internet has officially found its new favorite absurd test: ask 14 artificial intelligence image-makers to draw a frog with a Habsburg jaw and watch the chaos unfold. On paper, it sounds niche. In the comments, it turned into a full-on drama about whether these systems are clever, clueless, or just accidentally obsessed with monarchy. One sharp-eyed commenter loved the benchmark because it tests a very specific facial trait, then pointed out the wild twist: half the models seemingly smuggled “royalty” into the picture even though the prompt only asked for a jaw shape. Suddenly, the frogs weren’t just frogs, they were tiny cursed aristocrats.

That sparked the classic comment-section split. Some people were impressed that several models at least understood “Habsburg jaw” meant a dramatic underbite. Others absolutely roasted the results, with one brutal verdict saying none of them could be mistaken for human art. The harshest dunk was that many models seemed to know a big jaw was needed, but then slapped on a random blob that didn’t really connect to the frog’s face in any sensible way. Ouch.

And yes, the history lesson arrived right on cue: commenters explained that the Habsburg jaw is a real inherited facial condition associated with the famously inbred royal family, which made the accidental crowns and regal vibes even funnier—and weirder. The mood was basically half art critique, half meme lab, with one person already demanding even more contestants. Because of course the community saw 42 frog faces and immediately said: release another model into the arena.

Key Points

  • The article presents a personal benchmark based on a single prompt: "Generate an SVG of a frog with a Habsburg jaw."
  • The benchmark covers 14 models, with three tries per model per month, for a total of 42 runs in August 2026.
  • The page reports that all 42 runs successfully produced SVG output.
  • Displayed examples include Anthropic models Claude Opus 5, Claude Sonnet 5, Claude Haiku 4.5, and OpenAI GPT-5.5, along with run time and file size metadata.
  • The article analyzes whether SVG annotations stay structural or add interpretive language, such as anatomical exaggeration or royal framing.

Hottest takes

"Seven of fourteen models silently imported royalty into a prompt that named only an anatomical feature" — thebigship
"None of these could be remotely mistaken for human art" — getnormality
"They had some type of big blob for the jaw" — hn_throwaway_99
Made with <3 by @siedrix and @shesho from CDMX. Powered by Forge&Hive.