← All metrics
Largest training compute
Latest
~1e26 FLOP
2025
Time series
No verified measurements yet.
Definition The raw compute (floating-point operations) used to train the largest known model — the clearest single number behind AI's scaling, tracked by Epoch AI. It has grown ~4–5× per year; GPT-4 (2023) was the first at 1e25 FLOP, and frontier models passed 1e26 FLOP in 2025.
Source ↗By actor
50,000,000,000,000,000,000,000,000
38,000,000,000,000,000,000,000,000
21,000,000,000,000,000,000,000,000
3,400,000,000,000,000,000,000,000