Models
Frontier model
The largest, most capable general-purpose AI models at the cutting edge (GPT-5, Claude Opus, Gemini); the focus of frontier safety and regulation debates.
A frontier model is the name for the largest, most capable general-purpose AI models at the cutting edge of what is technically achievable — models that, at the moment of release, define the upper bound of attainable capability. Typical examples are GPT-5, Claude Opus, and Gemini.
The term borrows from the economic idea of a “technological frontier” and is deliberately relative: the frontier is not a fixed line but a moving one, so today’s frontier model becomes tomorrow’s baseline against which new systems are measured. Unlike the broader notion of a foundation model, “frontier” stresses position at the top of the capability curve and the potential for dangerous capabilities (offensive cyber operations, biological weapon design, autonomous behaviour that evades human control).
For that reason the term sits at the centre of AI safety and regulation debates. Leading labs publish “frontier safety frameworks” with dangerous-capability thresholds that, once crossed, trigger heightened safeguards. The EU AI Act and national AI safety institutes often draw the frontier category using compute thresholds (FLOPs), seeking an objective legal boundary.