Compare models available by June and December each year, then the latest available field. Higher points have stronger Arena text-preference scores; points further right have lower listed token prices. Press Play or choose a date to follow the frontier.
Historical reconstruction. The 2024–June 2026 scenes use ratings and prices captured later. They illustrate the changing model field; they are not exact dated leaderboards. The September scene selectively uses Arena’s 25 September 2026 leaderboard, checked on 30 September. Arena scores measure response preference, not professional-task performance.
Showing all labs. The frontier compares all shown models.