Frontier models tested in real life: egg capsules and bridges

Blogger Roberto Nickson forced frontier models to design 3D-printable structures in his AGENT WARS series and tested them in real life: in episode 03, only Claude Opus 5.5 successfully protected an egg, while in episode 02, its bridge held approximately 59 kg.

Compiled by Sergey KostenchukPublished 2026-10-19Updated 2026-10-19

2026-10-19 Research OpenAI

🤖 Frontier models tested in real life: egg capsules and bridges

Blogger Roberto Nickson hosts the AGENT WARS series: models design a structure themselves, he prints it on a 3D printer, and tests it. In episode 03, the egg survived the drop only with Claude Opus 5.5—from a height of 6 meters. Claude also won episode 02 with bridges: ~59 kg load versus 12 kg for the nearest competitor.

🌍 Such tests verify what is not visible in leaderboards: an agent's ability to plan a real structure under material limits and deliver a working product. The gap is huge: from a 59 kg load to a bridge that could not be assembled.

👤 Both episodes are available as short vertical videos with translation—you can see how differently models solve the same engineering task. This is useful if you are choosing a model for CAD or 3D printing. Episode 04 has already been announced.

Source 1: https://x.com/rpnickson/status/2111144364797817262 Source 2: https://x.com/rpnickson/status/2104234974350111108

Sources