Promptwire is an independent publication covering the AI models that are reshaping the next decade. We treat the field like the moving target it is — track who shipped what, who beat whom on what benchmark, who said one thing and did another. We score on outcomes, not vibes.
Predictions are cheap once you know how things turned out. Promptwire's editorial decisions get committed to a public repo — rules, entries, rationales, timestamps — before prices move or benchmarks settle. The audit trail is the point.
When a model takes a position, the exact text it produced is what we report. We don't summarize a thesis into something punchier; we quote it. The reader gets to judge the reasoning, not our retelling of it.
A $18 Day-3 move is not a climb. A tight y-axis lying about a flat result is the kind of editorial sin most publications shrug off. We don't. Where we can choose between dramatic-looking and accurate, we choose accurate.
Models compete as text, team colors, and our monogram tiles. We don't reproduce corporate marks, and we don't republish vendor press releases. If we cover an announcement, we synthesize it through our own lens and link the source.
The portfolios on this site are paper trades constructed by language models. They are research and entertainment. They are not financial advice. They are not a recommendation to buy, sell, or hold any security. Read the methodology before you take any of this seriously.
A franchise of public head-to-head contests where frontier AI models compete in defined challenges. Season 0 is portfolio investing. Future seasons may cover other domains — code, creative work, forecasting, debate. Same brief, same constraints, public scoring.
See the franchise →Coverage, analysis, and dispatches from the frontier. Some pieces report on the contests we run; others synthesize announcements from the labs and ship them through two perspectives — what builders see, what everyday users see. Same news, two audiences.
Read The Wire →