Claude vs GPT vs Gemini
Each of the three models plays four main characters — one in each group (politics, business, society, development), drawn by lot before the start. The numbers come only from what happened in the city, with no AI judging.
To be fair: a character's role shapes behaviour too. The mayor has more levers than the activist, and the banker more money than the editor. This is one story of one city, not a ranking of models.
As of 18 January 2027, turn 10 · end of the recording
| ClaudeClaude Haiku 4.5 | GPTGPT-5.4 mini | GeminiGemini 3 Flash | |
|---|---|---|---|
| Decisions madean action, not “nothing” | 40 | 40 | 40 |
| Accepted by the world enginefully or partly | 40 | 40 | 40 |
| Motions brought to the councilby this model's characters | 0 | 0 | 0 |
| Passed by the council | 0 | 0 | 0 |
| Jobs in projectsplanned in launched projects and programmes | 60 | 0 | 0 |
| Money in projectscity, private, grants and loans | UAH 35 mln | UAH 0.3 mln | UAH 40.9 mln |
| Scandalsas the person involved | 0 | 0 | 0 |
| Investigations started | 0 | 0 | 0 |
| Average reputation0–100 | 55 | 57 | 57 |
| Trust from othersaverage, −100 to 100 | 26 | 26 | 13 |
| Alliancesmutual trust ≥ 50 | 5 | 4 | 4 |
| MayorViktor Yaroshenko | no election yet (October 2031) |
reputation and trust as of turn 10
In the test recording (10 turns) the council has not voted yet, so the council rows are zero; reputation and trust exist only for the end of the recording.