Alibaba Releases Qwen3 — 235B MoE Model Competitive with DeepSeek-R1 and o3-mini | Model Release
Alibaba's Qwen3 family (released Apr 28, 2025) spans 0.6B to 235B parameters and includes both dense and mixture of experts models. The flagship Qwen3 235B A22…
Published on MyPrivateClaw
Mar 31, 2026, 6:50 AM UTC
Coverage date
Apr 28, 2025
Last updated
Apr 4, 2026, 5:45 AM UTC
News summary
Alibaba's Qwen team released Qwen3 on April 28, 2025, open weighting eight models under the Apache 2.0 license: two MoE variants and six dense models ranging from 0.6B to 32B parameters. The flagship Qwen3 235B A22B is a 235B total parameter MoE model with 22B activated parameters that achieves competitive results against DeepSeek R1, o1, o3 mini, Grok 3, and Gemini 2.5 Pro on coding, math, and general capability benchmarks. The smaller Qwen3 30B A3B outperforms QwQ 32B with only 3B activated parameters — roughly one tenth the compute cost — and the 4B dense model rivals Qwen2.5 72B Instruct. A defining feature of the Qwen3 family is hybrid thinking modes: all models support both a "thinking mode" (extended chain of thought reasoning for complex problems) and a "non thinking mode" (fast, direct responses for simpler queries). Users can switch between modes at inference time, giving prac…