Compute Threshold
A compute threshold is a level of training compute above which an AI model becomes subject to extra regulatory obligations.
What is Compute Threshold?
The EU AI Act uses 10^25 FLOPs to presume systemic risk, and California's SB 53 defines frontier models at 10^26 operations. Thresholds are simple to apply but imperfect, since efficiency gains let smaller models reach high capability; the EU Commission can update the figure.
How it works
Developers estimate their training compute and notify regulators when a threshold is reached or expected.
← All Artificial Intelligence & Superintelligence definitions · ← Back to the A–Z definitions