SI USA Superintelligence in the USA Subscribe
Menu

Timeline Compute SIU-LDG-2025-016 SI Index +2.00

OpenAI releases GPT-4.5, trained on an estimated 3.8 × 10²⁶ FLOP

OpenAI's largest pretrained model at release. Epoch AI estimates its training run at about 3.8 × 10²⁶ floating-point operations, the first documented run past 3 × 10²⁶.

Chart of estimated training compute for notable AI models from 2023 to early 2025, with GPT-4.5 the highest at 3.8 times 10 to the 26 FLOP
Estimated training compute of notable models released through February 2025. Chart: SI USA, from Epoch AI data / Data CC BY 4.0 · License

OpenAI described GPT-4.5 as its largest model and released it as a research preview. Epoch AI’s database records xAI’s Grok 3 (about 3.5 × 10²⁶) and, in July 2025, Grok 4 (about 5 × 10²⁶) as the next-largest documented runs.

Index treatment. Training run of 10²⁷ FLOP moves to 0.5.