Qwen3

Overview

This is a model family overview. For version-specific details, see the individual model entries linked below.

Qwen3 is Alibaba’s Qwen generation built around hybrid thinking modes, multilingual reasoning, and strong coding performance. The original April 29, 2025 release centered on a broad set of open-weight models under Apache 2.0, and Alibaba has since extended the family through Qwen3.5, the hosted Qwen3-Max tier, and the April 2026 Qwen 3.6 generation that adds frontier hosted previews and new open-weight variants.

Current Latest

Qwen 3.6 Max Preview is Alibaba’s current hosted flagship in the Qwen3 family, released in April 2026 with a sparse Mixture-of-Experts architecture, integrated thinking mode, and a 262K-token context window. The April 2026 wave also added open-weight Qwen3.6-27B and Qwen 3.6 35B-A3B variants on Hugging Face for self-hosted deployments. The earlier April 2025 open-weight releases, Qwen3-235B-A22B and Qwen3-30B-A3B, plus the hosted Qwen3-Max and Qwen3-Coder-Plus tiers, remain available alongside the new 3.6 variants.

Strengths

• Strong multilingual performance across 119 languages and dialects.
• Hybrid thinking and non-thinking modes in the core Qwen3 release.
• Broad family coverage from open-weight variants to premium hosted models.
• Good reasoning and coding performance for production assistants.
• Tool-use support for agentic orchestration workflows.
• Flexible deployment paths through Alibaba Cloud, Ollama, LM Studio, and other open-model ecosystems.

When to Choose Qwen3

Choose Qwen3 when multilingual quality and regional deployment flexibility are first-order requirements. The family is also a good fit when you want to mix open models for controlled environments with hosted premium tiers for harder tasks.

Access

• Alibaba Cloud DashScope / Model Studio APIs
• Official Qwen open releases on Hugging Face, ModelScope, and Kaggle
• Third-party inference and deployment partners