Olmo 3

Training compute

1.1×10²⁴ FLOP

Parameters

32B

Published

Nov 20, 2025

Olmo 3 is an AI model developed by Allen Institute for AI (United States), first published in November 2025. It works in the language domain, on tasks such as language modeling/generation, chat, reasoning and code generation.

Training it took an estimated 1.1×10²⁴ FLOP of compute. The model has 32,000,000,000 parameters. It was trained on roughly 5.5T datapoints. Training ran on 1,024 NVIDIA H100 SXM5 80GB for about 1.3K hours.

Access: Open weights (unrestricted). Its weights are openly available. Epoch AI rates the confidence of this record as confident.

Full record

Organization: Allen Institute for AI
Country of organization: United States
Domain: Language
Task: Language modeling/generation, Chat, Reasoning, Code generation
Training compute: 1.1×10²⁴ FLOP
Parameters: 32,000,000,000
Dataset size: 5.5T
Training hardware: NVIDIA H100 SXM5 80GB
Chips used: 1,024
Training time: 1,344 h
Training power draw: 1.4 MW
Model accessibility: Open weights (unrestricted)
Open weights: Yes
Epoch confidence: Confident