Qwen3-Max

Training compute

1.5×10²⁵ FLOP

Parameters

1T

Published

Sep 5, 2025

Qwen3-Max is an AI model developed by Alibaba (China), first published in September 2025. It works in the language domain, on tasks such as language modeling/generation, question answering, mathematical reasoning and 4 more.

Training it took an estimated 1.5×10²⁵ FLOP of compute (estimation method: operation counting). The model has 1,000,000,000,000 parameters. It was trained on roughly 36T datapoints.

Access: API access. Its weights are not openly released. Epoch AI rates the confidence of this record as speculative.

Full record

Organization: Alibaba
Country of organization: China
Domain: Language
Task: Language modeling/generation, Question answering, Mathematical reasoning, Code generation, Quantitative reasoning, Retrieval-augmented generation, Translation
Training compute: 1.5×10²⁵ FLOP
Compute estimation method: Operation counting
Parameters: 1,000,000,000,000
Dataset size: 36T
Model accessibility: API access
Open weights: No
Epoch confidence: Speculative

Benchmark results

01

Epoch Capabilities Index

ECI Score

143.8