Open WeightAlibaba

Qwen 3

Context

128K tokens

Modalities

text, code

Released

Apr 2025

Overview
Alibaba's flagship 235B MoE language model with exceptional multilingual capabilities across 100+ languages. Qwen 3 offers both thinking and non-thinking modes, with open weights available for the full model family.
Why it matters
Qwen 3 is the strongest evidence that the US-China AI competition is producing multiple frontier-capable model families. Its 235B MoE architecture with only 22B active parameters achieves excellent efficiency, and its multilingual strength across 100+ languages — especially CJK languages — makes it the go-to choice for applications targeting Asian markets. For multinational enterprises, Qwen's multilingual RAG capability is notably strong. Alibaba's cloud distribution through Aliyun gives it natural reach across Asia-Pacific, making it a serious alternative to Western models for non-English-primary deployments.

Key strengths

  • 235B MoE with only 22B active — excellent efficiency
  • Best-in-class multilingual support (100+ languages)
  • Dual thinking/non-thinking modes
  • Strong performance on agentic and tool-use benchmarks
  • Full model family open-weight (0.6B to 235B)

We cover ai models every week.

Get the 5 AI stories that matter — free, every Friday.

Know the terms. Know the moves.

Get the 5 AI stories that matter every Friday — free.

Free forever. No spam.