Ring-flash-linear-2.0

Training compute

7.7×10²³ FLOP

Parameters

104.2B

Published

Oct 23, 2025

Ring-flash-linear-2.0 is an AI model developed by Ant Group (China), first published in October 2025. It works in the language domain, on tasks such as language modeling/generation, question answering, quantitative reasoning and 2 more.

Training it took an estimated 7.7×10²³ FLOP of compute (estimation method: operation counting). The model has 104,200,000,000 parameters. It was trained on roughly 1T datapoints. Training ran on 32 NVIDIA H800 SXM5.

Access: Open weights (unrestricted). Its weights are openly available. It is built on top of Ling-flash-base-2.0-20T. Epoch AI rates the confidence of this record as confident.

Full record

Organization: Ant Group
Country of organization: China
Domain: Language
Task: Language modeling/generation, Question answering, Quantitative reasoning, Code generation, Mathematical reasoning
Training compute: 7.7×10²³ FLOP
Compute estimation method: Operation counting
Parameters: 104,200,000,000
Dataset size: 1T
Training hardware: NVIDIA H800 SXM5
Chips used: 32
Training power draw: 43.7 kW
Numerical format: FP8
Model accessibility: Open weights (unrestricted)
Open weights: Yes
Base model: Ling-flash-base-2.0-20T
Epoch confidence: Confident