Codex

Training compute

7.3×10²² FLOP

Parameters

12B

Published

Jul 7, 2021

Codex is an AI model developed by OpenAI (United States), first published in July 2021. It works in the language domain, on tasks such as code autocompletion.

Training it took an estimated 7.3×10²² FLOP of compute. The model has 12,000,000,000 parameters. It was trained on roughly 52.8B datapoints.

Access: API access. Its weights are not openly released. It is built on top of GPT-3 13B. The reference paper has 9,577 citations. Epoch AI rates the confidence of this record as likely.

Full record

Organization: OpenAI
Country of organization: United States
Domain: Language
Task: Code autocompletion
Training compute: 7.3×10²² FLOP
Parameters: 12,000,000,000
Dataset size: 52.8B
Model accessibility: API access
Open weights: No
Base model: GPT-3 13B
Citations: 9,577
Epoch confidence: Likely