En vivo
AI models

UL2

Training compute
1.2×10²³ FLOP
Parameters
20B
Published
May 10, 2022

UL2 is an AI model developed by Google Research and Google Brain (United States), first published in May 2022. It works in the language domain, on tasks such as language modeling/generation, question answering and text summarization.

Training it took an estimated 1.2×10²³ FLOP of compute (estimation method: hardware,operation counting). The model has 20,000,000,000 parameters. It was trained on roughly 1T datapoints. Training ran on 512 Google TPU v4 for about 744 hours. The compute alone is estimated at $127K in 2023 dollars.

Access: Open weights (unrestricted). Its weights are openly available. The reference paper has 387 citations. Epoch AI rates the confidence of this record as confident.

Full record
Organization
Google Research, Google Brain
Country of organization
United States
Domain
Language
Task
Language modeling/generation, Question answering, Text summarization
Training compute
1.2×10²³ FLOP
Compute estimation method
Hardware, Operation counting
Parameters
20,000,000,000
Dataset size
1T
Training hardware
Google TPU v4
Chips used
512
Training time
744 h
Chip-hours
380.9K
Training power draw
349.6 kW
Training cost (2023 USD)
$127K
Numerical format
BF16
Model accessibility
Open weights (unrestricted)
Open weights
Yes
Citations
387
Epoch confidence
Confident
More from Google Research,Google Brain
SourceEpoch AI, 'AI Models'. Published online at epoch.ai. Retrieved 2026-07-29 from https://epoch.ai/data/ai-models. Licensed under CC BY 4.0.
← All ai models