Qwen3-235B-A22B

Training compute

4.8×10²⁴ FLOP

Parameters

235B

Published

Apr 29, 2025

Qwen3-235B-A22B is an AI model developed by Alibaba (China), first published in April 2025. It works in the language domain, on tasks such as language modeling/generation, question answering, mathematical reasoning and 3 more.

Training it took an estimated 4.8×10²⁴ FLOP of compute (estimation method: operation counting). The model has 235,000,000,000 parameters. It was trained on roughly 36T datapoints.

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

Full record

Organization: Alibaba
Country of organization: China
Domain: Language
Task: Language modeling/generation, Question answering, Mathematical reasoning, Quantitative reasoning, Code generation, Translation
Training compute: 4.8×10²⁴ FLOP
Compute estimation method: Operation counting
Parameters: 235,000,000,000
Dataset size: 36T
Model accessibility: Open weights (unrestricted)
Open weights: Yes
Epoch confidence: Likely

Benchmark results

01	Epoch Capabilities Index	ECI Score	139.6
02	ForecastBench	Overall score	59.7
03	Aider Polyglot	Percent correct	59.6
04	Lech Mazur Writing	Mean score	8.3
05	GPQA Diamond	mean_score	0.71
06	MATH Level 5	mean_score	0.69
07	WeirdML	Accuracy	0.37
08	SimpleBench	Score (AVG@5)	0.31