A frontier model is a general-purpose AI model at the leading edge of what AI can do. There is no single legal definition, so lawmakers often use the amount of computing used in training as a stand-in: California’s SB 53 covers models trained with more than 10^26 operations.
You meet the term in AI safety rules and pledges, such as Australia’s planned obligations for frontier developers and the Paris declaration of November 2026, which focus testing and incident reporting on this small group of models.