OpenAI Five

Training compute

6.7×10²² FLOP

Parameters

159M

Published

Dec 13, 2019

OpenAI Five is an AI model developed by OpenAI (United States), first published in December 2019. It works in the games domain, on tasks such as dota 2. It counts among the frontier models: the systems trained with the most compute of their moment.

Training it took an estimated 6.7×10²² FLOP of compute (estimation method: reported). The model has 159,000,000 parameters. It was trained on roughly 454.2B datapoints. Training ran on 1,536 NVIDIA P100 for about 7.1K hours. The compute alone is estimated at $4M in 2023 dollars.

Access: Unreleased. Its weights are not openly released. The reference paper has 2,137 citations. Epoch AI rates the confidence of this record as confident.

Full record

Organization: OpenAI
Country of organization: United States
Domain: Games
Task: Dota 2
Training compute: 6.7×10²² FLOP
Compute estimation method: Reported
Parameters: 159,000,000
Dataset size: 454.2B
Training hardware: NVIDIA P100
Chips used: 1,536
Training time: 7,104 h
Chip-hours: 10.9M
Training power draw: 786.4 kW
Training cost (2023 USD): $4M
Model accessibility: Unreleased
Open weights: No
Citations: 2,137
Epoch confidence: Confident