AI Power Rankings
Fifteen frontier AI labs, ranked across ten dimensions by a panel of AI evaluator models. The structural-power index of who actually leads AI.
Background
Most claims about who is "winning" in AI come from a single voice: one analyst, one lab's own marketing, one headline. I wanted a ranking that checks itself, so I built one that asks several frontier AI models the same question and compares their answers.
How it works
Fifteen AI companies are scored across ten evaluation areas, from model quality and research to safety, talent, and distribution. Each evaluator model rates every company on every area, has to name the counterargument before committing to a score, and is asked directly to check its own bias toward the lab that built it. Scores are averaged across models to cancel out individual bias, and every score carries a written justification.
Technology
The scoring pipeline is Python and SQLite: a self-contained scoring prompt that any model can run, JSON outputs imported into a database, and command-line tools to query the rankings overall, by area, by company, or over time. The public site presents that same dataset.
How it was built
I wrote the methodology and scoring rubric myself, then handed the identical instructions to different AI models so the disagreements between them become part of the signal, not noise to hide. Like my other projects, I built it spec-first with AI tools rather than writing it by hand, and the full methodology is open so anyone can run their own evaluation and compare notes.