GDPx: the benchmark where every model fails
SemiAnalysis introduces GDPx, a frontier-model capability index scored against real economic tasks. Top models reach 12% — a number that defines the gap between capability claims and economic deployability.
- GDPx measures AI on tasks representing ~10% of global GDP
- Best frontier models score 12%; GPT-4-class achieves ~4%
- Built to track the gap between marketing claims and real-world capability