Llama Nemotron Ultra 253B

Training compute

3.9×10²⁵ FLOP

Parameters

253B

Published

Mar 18, 2025

Llama Nemotron Ultra 253B is an AI model developed by NVIDIA (United States), first published in March 2025. It works in the language domain, on tasks such as language modeling/generation, question answering, quantitative reasoning and 2 more. It counts among the frontier models: the systems trained with the most compute of their moment.

Training it took an estimated 3.9×10²⁵ FLOP of compute (estimation method: operation counting). The model has 253,000,000,000 parameters. It was trained on roughly 603B datapoints.

Access: Open weights (restricted use). Its weights are openly available. It is built on top of Llama 3.1-405B. Epoch AI rates the confidence of this record as likely.

Full record

Organization: NVIDIA
Country of organization: United States
Domain: Language
Task: Language modeling/generation, Question answering, Quantitative reasoning, Code generation, Neural Architecture Search - NAS
Training compute: 3.9×10²⁵ FLOP
Compute estimation method: Operation counting
Parameters: 253,000,000,000
Dataset size: 603B
Model accessibility: Open weights (restricted use)
Open weights: Yes
Base model: Llama 3.1-405B
Epoch confidence: Likely