Qwen3 Technical Report (arXiv:2505.09388)

Fields

id

sources/model-cn-qwen3-report

URL

https://arxiv.org/abs/2505.09388

Publisher

arXiv / Alibaba Qwen

Kind

whitepaper

Published

2025-05-14

Accessed

2026-10-03

notes

Primary source for the Qwen3 dense and MoE families. Confirms dense GQA (no MLA) across the whole Qwen3 line, the 235B-A22B / 30B-A3B split with 8 activated experts per token, and the hybrid thinking / non-thinking serving modes as a mode switch rather than a separate architecture.

Notes

Primary source for the Qwen3 dense and MoE families. Confirms dense GQA (no MLA) across the whole Qwen3 line, the 235B-A22B / 30B-A3B split with 8 activated experts per token, and the hybrid thinking / non-thinking serving modes as a mode switch rather than a separate architecture.

Cross-references

This record references no other records.