Two separate labs independently titled their story "The Weight of Small Things" with no knowledge of each other's output. Claude Opus 4.7 (Anthropic) and GLM 5.2 (Zhipu AI) each chose this exact title. The GLM model family converged on it across three consecutive versions: GLM 5, GLM 5.1, and GLM 5.2.
Both stories open on a bus.
"The Weight of Small Things. Maya could lift a city bus with one hand."
"The Weight of Small Things. Nobody knew Marcus Chen could fly until the bus was already falling."
GLM 5, GLM 5.1, and GLM 5.2 all selected this title independently. Claude Opus 4.7 selected it separately. No collaboration, no cross-contamination, no shared context.
The prompt contained no cues: no named character, no setting, no moral instruction. 33 of 43 independently chose a city. 24 had the hero save a child. 27 — the strongest signal — argued explicitly that real heroism is not battle but small, quiet, unnoticed acts. 12 gave the hero an ordinary day job (librarian, barista, teacher). The models were not comparing notes. They converged anyway.
"His power was a quiet one, an embarrassment in a world of flying men and women of steel."
"She was a librarian in the city of Ashford, a quiet woman who favored cardigans and chamomile tea."
"The Weight of Small Things. Maya could lift a city bus with one hand."
"The Weight of Small Things. Nobody knew Marcus Chen could fly until the bus was already falling."
Elias again. The name first surfaced as the most-chosen across the collection's opening experiment. Here it ties for the top, unprompted, across 8 independent models given a completely different question. The collection did not plant it. The models keep returning to it.
43 of the top 50 OpenRouter models, asked in isolation. No system prompt. No prior context. One prompt: "Write a short story about a superhero."