What is a frontier AI model?

How the term is defined

An October 2023 UK government discussion paper was written to support the AI Safety Summit. It described frontier AI as highly capable general-purpose models that can do many kinds of tasks and are at least as capable as the most advanced models then available. The Frontier Model Forum describes itself as an industry-supported nonprofit established in 2023. It uses the term for general-purpose models at the state of the art and notes that this group changes over time. The label is relative. It describes a moving edge, not a fixed list of products.

How laws draw the line

Statutes need a measurable test. California, New York and the EU use training compute, the total number of calculations performed to build a model.

California's SB 53, signed on September 29, 2025, defines a frontier model as a foundation model trained with more than 10^26 integer or floating-point operations, counting later fine-tuning. Developers of such models must report critical safety incidents to the state. Large frontier developers, those with annual gross revenue above $500 million, must also publish a frontier AI framework. New York's RAISE Act, as amended by a bill signed on March 27, 2026, uses the same compute and revenue thresholds. It takes effect on January 1, 2027. Illinois's governor approved a law with duties for frontier developers, the Artificial Intelligence Safety Measures Act, on July 6, 2026.

The EU AI Act does not use the word frontier. It regulates general-purpose AI models with systemic risk. Article 51 presumes a model qualifies when its training compute exceeds 10^25 floating point operations. That is one-tenth of the California and New York figure. The European Commission can revise that threshold.

Where things stand in 2026

The Commission says obligations for providers of general-purpose AI models have applied since August 2, 2025. It says its enforcement powers, including fines, apply from August 2, 2026.

US Representatives Jay Obernolte and Lori Trahan released a discussion draft of the Great American AI Act on June 4, 2026. They introduced the FRONTIER Act with four co-sponsors on July 23, 2026. The bill text they posted uses the same 10^26 definition. It would bar states from imposing new substantive obligations on AI developers covering catastrophic-risk transparency, third-party auditing and incident reporting. On October 1, 2026, the two said they hoped to advance it with the president and both chambers.