Qwen3 Technical Report

Alibaba's Qwen3 family - dense + MoE models with hybrid reasoning, 119-language coverage, and strong agentic / coding / math performance.

Open

Publisher: Alibaba Qwen (Tongyi Qianwen)
Year: 2025
Venue: preprint
ArXiv: arxiv.org/abs/2505.09388
URL: arxiv.org/abs/2505.09388
Code: github.com/QwenLM/Qwen3
Authors: 60
Hosting: External sourcelicense unknown

Cite

Notes

Only stored in your browser.

Attribution

Abstract & full text: arxiv.org/abs/2505.09388
TL;DR: semanticscholar.org/paper/d2d84d56f730f81d276a02b48d5d44db5bde0b4a
Code: github.com/QwenLM/Qwen3

Attribution policy →

Introduces 2 artifacts - 2 models

TL;DR

Semantic Scholar

Empirical evaluations demonstrate that Qwen3 achieves state-of-the-art results across diverse benchmarks, including tasks in code generation, mathematical reasoning, agent tasks, etc., competitive against larger MoE models and proprietary models.

Artifacts

Models

Qwen3-235B-A22B Qwen3 (family)