Frontier
Every model, plotted the moment it shipped — then priced to run.
90d → 12m · quiet quarter
24/31 scored on GPQA Diamond · 7 in the unscored band.Saturated at the frontier (>90%); Anthropic announced it will stop reporting it. methodology.
The full read
2,409 models · 15 labs · 31 featured
62 scored on GPQA Diamond · 22 API providers · rentals from 0 marketplaces
Updated —
Click any model to price it to run — self-host, rental, and API in one place.
score viewopen / closedsize = host difficultyapprox. dateOne benchmark or a versioned percentile composite on Y — never a fabricated index. How Frontier works.