Pangu Ultra MoE

Training compute

3.1×10²⁴ FLOP

Parameters

718B

Published

May 7, 2025

Pangu Ultra MoE is an AI model developed by Huawei (China), first published in May 2025. It works in the language domain, on tasks such as language modeling/generation and question answering.

Training it took an estimated 3.1×10²⁴ FLOP of compute (estimation method: operation counting). The model has 718,000,000,000 parameters. Training ran on 6,000 Huawei Ascend 910B.

Access: Hosted access (no API). Its weights are not openly released. Epoch AI rates the confidence of this record as speculative.

Full record

Organization: Huawei
Country of organization: China
Domain: Language
Task: Language modeling/generation, Question answering
Training compute: 3.1×10²⁴ FLOP
Compute estimation method: Operation counting
Parameters: 718,000,000,000
Training hardware: Huawei Ascend 910B
Chips used: 6,000
Training power draw: 4.7 MW
Model accessibility: Hosted access (no API)
Open weights: No
Epoch confidence: Speculative