GroveAI
AI Profile

Qwen 3: Alibaba's Open-Weight Powerhouse

Qwen 3 is Alibaba Cloud's latest open-weight model family featuring hybrid thinking/non-thinking modes, sizes from 0.6B to 235B (MoE), and strong multilingual, coding, and reasoning capabilities.

Specifications

At a glance

Parameters

0.6B / 1.7B / 4B / 8B / 14B / 30B / 32B / 235B-A22B (MoE)

Context Window

128,000 tokens

Training Data Cutoff

Early 2025

Release Date

April 2025

Licence

Apache 2.0

Pricing

Free (self-hosted) or via Alibaba Cloud

Overview

About Qwen 3

Qwen 3 is Alibaba Cloud's third-generation large language model family, marking a significant leap forward with hybrid thinking modes, broader model sizes, and improved performance across the board. The flagship Qwen 3 235B-A22B uses a Mixture of Experts architecture, activating only 22B parameters per token from a total of 235B, delivering frontier-competitive performance with efficient inference. The standout feature of Qwen 3 is its hybrid thinking mode: models can dynamically switch between a fast 'non-thinking' mode for straightforward queries and a slower 'thinking' mode with extended chain-of-thought reasoning for complex problems. This makes Qwen 3 competitive with dedicated reasoning models while remaining practical for general-purpose use. Released under the permissive Apache 2.0 licence across all variants, Qwen 3 has become one of the most popular open-weight model families globally. It supports over 100 languages, with particular strength in Chinese, English, and other Asian languages. The range of sizes — from 0.6B for edge deployment to 235B MoE for frontier tasks — makes it one of the most versatile open model families available.

Strengths

Capabilities

  • Hybrid thinking/non-thinking modes for adaptive reasoning
  • Wide range of sizes from 0.6B to 235B (MoE) parameters
  • 100+ language support with broad multilingual capabilities
  • 128K context window across model sizes
  • Apache 2.0 licence for all variants enabling unrestricted commercial use
  • Strong mathematical, coding, and scientific reasoning
  • MoE flagship for efficient frontier-class inference

Considerations

Limitations

  • MoE flagship model requires significant infrastructure to self-host
  • Ecosystem still smaller than Llama's in Western markets
  • Western language performance can trail Llama 4 on some tasks
  • Documentation quality varies between Chinese and English

Best For

Ideal use cases

  • Bilingual Chinese-English applications
  • Cost-effective self-hosted deployments with permissive licensing
  • Tasks requiring adaptive reasoning depth (hybrid thinking)
  • Mathematical reasoning and code generation tasks
  • Edge and mobile deployment with smaller model variants

Pricing

Free under Apache 2.0 for self-hosting. Also available through Alibaba Cloud Model Studio and various inference providers at competitive rates.

FAQ

Frequently asked questions

Qwen 3 models can switch between 'thinking' mode (extended chain-of-thought reasoning for complex problems) and 'non-thinking' mode (fast direct answers for simple queries). This allows a single model to handle both simple and complex tasks efficiently.

Both are competitive open-weight model families with MoE architectures. Qwen 3 offers hybrid thinking modes and Apache 2.0 licensing. Llama 4 has native multimodal support and Scout's massive 10M context window. Both are strong choices depending on your specific requirements.

Yes. All Qwen 3 variants are released under the Apache 2.0 licence, one of the most permissive open-source licences. There are no usage restrictions, user count limits, or commercial use restrictions.

The 235B MoE model activates only 22B parameters per token, making inference more efficient than a dense 235B model. It still requires multi-GPU setups (e.g., 4x A100 80GB). Smaller variants like the 8B and 14B models run comfortably on consumer hardware.

Qwen 3 is developed by Alibaba Cloud's Qwen team based in China. Alibaba's continued investment in open AI research has made Qwen one of the world's most popular open-weight model families.

Need help with Qwen 3?

Our team can help you evaluate and implement the right AI tools. Book a free strategy call.